An official website of the United States government
The .gov means it’s official. Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.
The site is secure. The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.
- Account settings
- Advanced Search
- Journal List
- Endocrinol Diabetes Metab
- v.3(3); 2020 Jul
A quantitative measure of treatment response in recent‐onset type 1 diabetes
Brian n. bundy.
1 Health Informatics Institute, University of South Florida, Tampa FL, USA
Jeffrey P. Krischer
All TrialNet data generated or analysed during this study can be requested from the NIDDK Central Repository at https://repository.niddk.nih.gov/studies/trialnet/ .
This paper develops a methodology and defines a measure that can be used to separate subjects that received an experimental therapy into those that benefitted from those that did not in recent‐onset type 1 diabetes. Benefit means a slowing (or arresting) the decline in beta‐cell function over time. The measure can be applied to comparing treatment arms from a clinical trial or to response at the individual level.
An analysis of covariance model was fitted to the 12‐month area under the curve C‐peptide following a 2‐hour mixed meal tolerance test from 492 individuals enrolled on five TrialNet studies of recent‐onset type 1 diabetes. Significant predictors in the model were age and C‐peptide at study entry. The observed minus the model‐based expected C‐peptide value (quantitative response, QR) is defined to reflect the effect of the therapy.
A comparison of the primary hypothesis test for each study included and a t test of the QR value by treatment group were comparable. The results were also confirmed for a new TrialNet study, independent of the set of studies used to derive the model. With our proposed analytical method and using QR as the end‐point, we conducted simulation studies, to estimate statistical power in detecting a biomarker that expresses differential treatment effect. The QR in its continuous form provided the greatest statistical power when compared to several ways of defining responder/non‐responder using various QR thresholds.
This paper illustrates the use of the QR, as a measure of the magnitude of treatment effect at the aggregate and subject‐level. We show that the QR distribution by treatment group provides a better sense of the treatment effect than simply giving the mean estimates. Using the QR in its continuous form is shown to have higher statistical power in comparison with dichotomized categorization.
This paper proposes a quantitative end‐point in the study of recent‐onset type 1 diabetes that measures the effect of a treatment on the stimulated C‐peptide of an individual patient and, in the aggregate, discriminate those who benefited (responders) from those who did not (non‐responders). The quantitative response measure can be used to evaluate promising biomarkers or other prognostic characteristics and is defined by the difference between the observed and expected C‐peptide levels.
Researchers have expressed a strong desire to define a measure of response at the individual level in studies of recent‐onset type 1 diabetes (T1D) subjects treated with an experimental agent from a completed randomized clinical trial. Motivation for this reflects a need to provide regulators, considering approval of an experimental therapy, a percentage of subjects that ‘benefited’ from that experimental therapy. As well, patients might better understand the potential benefit of treatment, if put in terms of the chance of achieving a specific state in their disease (eg reduced hypoglycaemic episodes). The desired state should be relevant and meaningful from the perspective of the patient. This paper addresses another purpose: to separate subjects that received the experimental therapy into those that benefitted from those that did not. Benefit for the recently diagnosed type 1 diabetes subject means a slowing (or arresting) the decline in beta‐cell function over time. Serum C‐peptide measured over 2 hours during a mixed meal tolerance test (MMTT) has become an accepted surrogate for beta‐cell function. Presumably, one might conclude that subjects that had minimal decline (or no decline) benefited from the therapy, while those with steeper declines did not. A counterargument is that the rate of decline varies independent of treatment and this variation may stem from differences in the underlying severity of the disease. Thus, it is unclear how to distinguish differential therapeutic benefit from disease severity in the treated group.
Additionally, a goal of biomarker studies is to identify a patient characteristic (usually genetic or immunological) that is associated with (and thus may explain) any variation in therapeutic effect. This association may help elucidate the mechanistic function of the therapy administered or identify a subpopulation for intervention as part of a strategy to refine and develop effective therapies. However, in general, a biomarker that is associated (or statistically speaking, correlated) with C‐peptide decline in the treated group is either predictive of benefit or prognostic (ie indicative of the severity of T1D). A prognostic variable (for recent‐onset T1D) is a characteristic, measured prior to therapy, which correlates with C‐peptide outcome. A key feature is that the correlation is present in both treated and untreated (ie placebo) groups. A predictive variable correlates with C‐peptide outcome only for the treated group, and no correlation exists in the placebo group. Thus, an initially promising biomarker requires testing in both placebo and treated samples to distinguish whether it is predictive or prognostic. Age is a classic example of a prognostic variable since there is a strong direct correlation between age and 1‐year C‐peptide decline regardless of the agents that TrialNet has studied to date. This does not preclude the possibility of age being predictive for some experimental agent in the future.
Many attempts at identifying subjects that have benefited from therapy (from those that have not benefited) have dichotomized the change (from baseline) in the stimulated C‐peptide from an MMTT. Responders are often defined to be those above some C‐peptide threshold and the complement being non‐responders. Herold et al 1 , evaluating the effect of anti–CD3‐based response on the change in C‐peptide level, defined as the area under the curve (AUC) mean increase over the fasting C‐peptide level. Response was considered when the value increased by more than 7.5% from the baseline value—7.5% was used because it is one‐half of the C‐peptide interassay coefficient of variation. Mortensen et al 2 used the coefficients from modelling C‐peptide regressing on HbA1c and insulin dose per kilogram weight to define a responder (if HbA1C per cent +4 ∙ insulin dose units per kilogram per 24 hours ≤ 9 then classify as responder). Again Herold et al 3 , evaluating the effect of anti‐CD20, defined response using the coefficient of variation estimate of 0.097. If the 6‐month C‐peptide AUC mean was greater or equal to 90.3 per cent of baseline (≤0.097 decrease), the subject was classified as a responder. In another report by Herold et al 4 , response was defined as <40% decline of C‐peptide at 2 years from baseline. This threshold was selected primarily because all control subjects had ≥40% decline. Beam et al 5 recommended using strictly no decrease in 6‐month C‐peptide from baseline to define responder. He indicated that the bias (‘the amount by which a responder definition will, on average, over‐ or underestimate the responder percentage in a patient population’) is nearly zero compared with definitions that include some percentage decline (eg 7.5%) as responders where the bias was not negligible. The limitations of these efforts appear ad hoc, possibly data‐driven and are applied to a single study.
This paper proposes the use of an adjusted end‐point we refer to as the quantitative response (QR) and a specific model structure (additive model with an interaction term) for screening biomarkers to determine their predictive or prognostic attribute. The basis of the QR is an analysis of covariance (ANCOVA) model of C‐peptide that adjusts for baseline C‐peptide and age and has been previously described. 6 The model forecasts the C‐peptide level at 12 months but does not include any effect of an active therapy. The QR is defined as the observed 12‐month C‐peptide AUC minus the model's predicted C‐peptide AUC (expected). This age‐adjusted value may reflect a differential benefit of treatment although distinguishing disease severity from a differential treatment effect on a subject‐by‐subject basis is impossible by simply viewing the QR distribution. It requires that a biomarker (eg expressing a mechanistic function of the therapy) be measured and analysed for any association (correlation) with the QR end‐point. The correlation needs to be quantified using an additive model (eg ANCOVA) with QR as the dependent variable and covariates: the biomarker, treatment group (0 = placebo and 1 = active therapy group) and the product of these two covariates (interaction term) as the third covariate. We demonstrate the advantage of the QR and the method of analysis as being general by making various use of six TrialNet studies, the method is not data‐driven (method was not based any biomarker and QR is independent of all but the first five recent‐onset TrialNet studies), and it is based on statistical method that allows expressing the prognostic and/or predictive feature of the biomarker being tested. Our hope is that this approach will provide a uniform and general framework for evaluating biomarkers when the goal is to determine whether the biomarker expresses differential treatment benefit (ie predictive biomarker).
2. MATERIALS AND METHODS
Baseline and one‐year follow‐up data from five completed TrialNet studies of recent‐onset type 1 diabetes subjects 7 , 8 , 9 , 10 , 11 were used in fitting an analysis of covariance (ANCOVA) model. The model cohort is also used to illustrate the utility of the quantitative response measure. The more recently completed antithymocyte globulin (ATG), with and without pegylated granulocyte colony‐stimulating factor (GCSF), trial 12 was also included to illustrate the generalizability of the proposed statistic on an independent data set. Participants completed a written informed consent and/or assent before participation in these studies. The eligibility for these studies was quite similar in that all had to meet the definition with respect to the diagnosis of type 1 diabetes and enrolment within 100 days of diagnosis and a C‐peptide level ≥ 0.2 pmol/mL. The studies did vary at the younger age range by design with an upper limit of 45 years.
2.2. Statistical considerations
The primary outcome is the C‐peptide levels from the first 2 hours of a mixed‐meal tolerance test (MMTT). The trapezoidal rule is applied to the five timed measurements and then summed to approximate the area under the curve then divided by the 120‐minute interval, henceforth C‐peptide AUC mean.
In our previous paper (Bundy & Krischer, 2016), we fit the ANCOVA model to the 1‐year C‐peptide AUC mean from the modelled cohort regressing on age at study entry, natural log‐transformed (after adding 1) baseline C‐peptide AUC mean and each experimental treatment assignment. Although there was no systematic process in considering other covariates, neither body mass index (transformed to z‐score) nor the second‐degree term for age (which allows a parabolic fit to transformed C‐peptide value) provided an improved model fit. HbA1c was statistically significant but because of collinearity with baseline C‐peptide the improvement in the model fit was negligible ( R 2 increased from .593 to .599). The following equation (from the fitted model) gives the predicted transformed 1‐year C‐peptide AUC mean given the age and baseline C‐peptide of any subject if administered placebo:
The Cp variables represent the pertinent C‐peptide AUC means, and Age is the year of age at study entry. ln is the natural logarithm function, and E [·] represents the expected value. The square root of the residual mean squared error (RMSE) is 0.151.
2.2.2. Quantitative response (QR)
Having measured the 1‐year C‐peptide AUC mean of a subject (usual units: nano‐moles per litre; the time units in minutes are cancelled out by division by 120 minutes), we can compute the transformed difference between their observed C‐peptide sAUC from their expected level (ie observed minus expected). Hence, for an individual, i , the QR is defined to be
(the units: plus‐one‐natural‐log of nano‐moles per litre). It may be useful to consider QR as an adjusted normalized value of the 1‐year C‐peptide with the empirical distribution centred at zero in the absence of any treatment effect.
To evaluate the variation of the QR as a function of the observed baseline C‐peptide (statistically referred to as heteroscedasticity), we followed White's method. 13 A linear regression model was fitted to the squared QR values regressing on baseline C‐peptide to estimate the change in variance by baseline C‐peptide. The fit indicated that the median square root of the RMSE was 0.152 (10th and 90th percentiles: 0.108 and 0.199). Although the variance of the QR varies with baseline C‐peptide, the QR is an accurate estimate (the statistical term is unbiased estimate. A non‐technical explanation is if the QR could be theoretically measured multiple times on the same subject, the average of those multiple values would be ever nearer the true QR value). The variance of QR was not correlated with age. (An analysis demonstrating that the QR is independent of age and baseline C‐peptide can be found in Appendix S1 : Figures S1 and S2 ).
2.2.3. Estimating statistical power
Monte Carlo simulation was employed to estimate the statistical power when testing for a correlation between a linear predictive variable (ie biomarker) and QR. Each simulation sampled baseline C‐peptide and age pairs from the modelled cohort at random with replacement (real data). The sample size for each simulated trial was set to typical size TrialNet phase II recent‐onset trial (ie 33 and 17 subjects in experimental and placebo group, respectively, having the end‐point of 1‐year C‐peptide AUC mean). The size was based on 50% minimal detectable difference (MDD; algebraically: Δ) in the treatment group means at 12 months poststudy entry (0.376 vs. 0.564 nano‐moles/L), 85% statistical power, type 1 error of 0.05 (1‐tail test), allocation ratio of 2:1 (experimental: placebo) and standard deviation of 0.158 (see Bundy & Krischer). The predictive biomarker was formulated such that it had a range from zero to twice the mean (to preserve symmetry) when following a normal distribution. The treatment effect varied linearly with the biomarker, that is, Δ∙ biomarker . For generalizability, two other distributions for the predictive variable were considered: the chi‐square and uniform distributions. The mean treatment effect for all three biomarker distributions was set at 1.18Δ. This value represents midway between the MDD and the largest treatment effect seen in a TrialNet study of recent‐onset type 1 diabetes (the ATG/GCSF study had an effect of 1.37Δ). 12 For the normal and uniform distributions, the treatment effect ranged from 0 to 2.36Δ. The right‐skewed square root of the chi‐square distribution function required extending the range to 4.25 and also setting the degrees of freedom to 1.815 in order to keep the mean the same as the other distributions. For the normal distribution, the standard deviation was set to 1.18/2; extreme values outside the range of (0, 2.36) were resampled for both the normal and the chi‐square distributions. To reflect biomarker measurement error (unexplained variation), an independent normally distributed random variables with mean 0 and variances of σ 2 and σ 2 /2 were added to the predicted value. The σ 2 is the unexplained variance of the QR (or C‐peptide AUC mean after adjustment) which was estimated as 0.151 from the modelled cohort.
The choice of the biomarker measurement error is somewhat arbitrary but considered to be in the same range as QR measurement error. The primary purpose is to illustrate the loss in power for some response definitions. We did not consider a dichotomized predictive variable for the sake of simplicity. In general, considerably larger sample size is required to have nominal power in detecting a dichotomous predictive variable. We set the simulations to 20 000 replications, which yields a maximum 95% confidence interval for the statistical power of ±0.00693 (binomial approximation).
A formal statistical test was employed to each simulated trial with a threshold of significance set at 0.05 (one‐sided). The test was based on a three covariate ANCOVA model of QR: biomarker, treatment group (parametrized: 0 = placebo group and 1 = experimental treatment group) and the product (interaction) term; the coefficient of this last term was the basis of the test.
2.2.4. Percentile‐based responder
A series of QR thresholds were used to illustrate a forced dichotomized responder definitions to be evaluated in the simulation. The goal was to determine what QR threshold might be preferable, if one required classifying subjects as responders and non‐responders on the basis of the QR. The control group was used to establish any shift in the QR distribution by calculating the mean and the average dispersion for each treatment group (pooled variance). The percentiles extracted from the normal distribution were used as thresholds to classify responders in the treated group. The logic of this algorithm is that it corrects the thresholds for a trial where the QR distribution is not centred at zero (this may happen when the eligibility criteria have been altered or the experimental agent being studied has altered the subject selection process).
All analyses were conducted in TIBCO Spotfire S+™ 8.2 Workbench.
The expected C‐peptide level at 1 year for a subject is calculated by substituting their age at entry, and their baseline C‐peptide value in Equation 1 . Figure 1 displays the observed transformed 1‐year C‐peptide and the corresponding expected C‐peptide for all placebo group subjects from the modelled cohort. The figure provides a visual of the observed minus expected distribution via the vertical distances from the diagonal line to each point (the QR is negative if the point is below the diagonal line). There are 60 (50.8%) positive and 58 negative QRs indicating symmetry around zero overall, and the symmetry is reasonably consistent across the range of expected baseline C‐peptide values. The variation of the QR was greater for greater values of expected C‐peptide. No such correlation was present for age. A regression line adequately expresses the variation in QR for the range of baseline C‐peptide. The median variation was 0.152, and the 10th and 90th percentiles were 0.108 and 0.199, respectively. Regardless of this change in the QR variation, the QR value is accurate (unbiased estimator).
Observed 1‐year C‐peptide by expected model‐based C‐peptide for subjects in the placebo groups. The vertical distance from the point to the diagonal line is the observed 1‐year C‐peptide minus the model‐based expected 1‐year C‐peptide, that is, the quantitative response (QR)
Figure 2 is a boxplot of the QRs for three treatment groups and displays all observations. The symmetry of QRs around zero is reaffirmed in the placebo group. In contrast, there is a positive shift in the QR distribution for the experimental treatment groups compared with the placebo group from these two trials (TrialNet rituximab and abatacept studies). 8 , 10 There are 34 (68.0%) and 48 (68.6%) subjects with QRs greater than zero in the rituximab‐ and abatacept‐treated groups, respectively. The median QR is an estimate of the treatment effect albeit on a scale which is not relatable clinically. We suggest that displaying the QRs for every subject by treatment group provides a better sense of the treatment effect than simply the average, specifically, that there may be a fair number of subjects in the treated group with lower QRs than the control group despite an overall positive result.
Box plot of quantitative response of three treatment groups: Placebo [from the combined studies], rituximab 8 and abatacept. 10 The horizontal sides of the box mark the 25th and 75th percentiles, the horizontal line within the box marks the median, and the whiskers mark the 5th and 95th percentiles. The horizontal variation of the points is for visual clarity and has no meaning
The QR may be used to test a treatment effect using a simple t test, (or a non‐parametric test such as the Wilcoxon rank‐sum test). Table 1 displays a comparison of the recomputed primary hypothesis test (any slight difference from published results is due to post‐publication data corrections) and a t test of the QR value by treatment group. The table includes the treatment effect coefficient from the analysis of covariance model adjusting for age and baseline C‐peptide and the Wald significance level (the stipulated primary hypothesis test used in the primary analysis of each study). For comparison, the mean difference in the QR between treatment groups and the significance level of the associated t test are provided. The right side of the table is the comparison of the second experimental treatment to the control group for the three TrialNet studies that randomized to three treatments. The two analytical approaches for each comparison agree. The TrialNet ATG/GCSF study 12 is independent of the data used in determining the QR coefficients [Equation 2 ]; yet, the two approaches agree for both experimental groups.
Treatment effect estimates and significance levels for six TrialNet studies by the original primary hypothesis test (Wald test for treatment from the ANCOVA model) and two‐sample t test of quantitative response (QR)
A histogram of the QRs for each treatment group from TrialNet ATG/GCSF study is in Figure 3 . Also displayed are the normal bell‐shaped curve (probability density) determined from the average QR of each group and an overall variation of QR (pooled variance). It is visually apparent that there is an increase in the QRs in the ATG‐only group compared with the placebo group and the treatment effect appears to be a positive shift for the entire group rather than for just certain levels of QR’s.
Quantitative response histogram by treatment group of TrialNet ATG/GCSF study: panel A is the placebo group, panel B is the ATG and GCSF group, and panel C is the ATG‐only group. Superimposed on each histogram is the normal density function where the mean was set to the mean of the quantitative response of the group and the standard deviation was set to the pooled (over the 3 groups) standard deviation of the quantitative response. Panel D displays just the three treatment group normal density functions using the same line type as in panels A‐C, respectively, solid = placebo, short dash = ATG only and long dash = ATG and GCSF
We propose an analytical method to screen biomarkers to assess their prognostic or predictive attribute. Fit a statistical model of QR (dependent variable) with three covariates (independent variables): the biomarker to be evaluated, the treatment group category (both treatment groups: 0 = placebo and 1 = treatment) and an interaction term of the biomarker and treatment covariates. The fitted coefficients and their significance levels will delineate the biomarker as prognostic, predictive, both or neither. Figure 4 provides scatterplots of QR and three hypothetical biomarkers representing three possible relationships with treatment and the end‐point QR. These hypothetical biomarkers represent prognostic, predictive and both in panels A, B and C, respectively. All three figures reflect an active experimental therapy. In addition, panels A and B display a positive association between the biomarker and QR. Panel C displays a negative association between the biomarker and treatment effect but a positive association between the biomarker and QR for the placebo group only. Other relationships are possible.
Scatterplots of the relationship of three types of hypothetical biomarkers with treatment group and QR. Panel A: prognostic biomarker, Panel B: predictive biomarker, Panel C: both prognostic and predictive biomarker. Other relationship other than these three may exist. The variation around the regression line (not shown) has been set low for visual clarity
We evaluated the chance of detecting (statistical power) a predictive biomarker using the model structure described above and a formal statistical test associated with the interaction term. The simulation studies set the treatment effect to be proportional to the value of the biomarker (linear effect). Measurement errors were included for both the QR (via C‐peptide AUC mean) and the biomarker to reflect the reality of measuring constituents from subject samples. Although we do not suggest strict adherence to 0.05 significance level when evaluating biomarkers (lack of statistical power is often an issue when testing interaction terms) for their predictive attribute, we did so for the simulations. Table 2 presents the chance of detecting (statistical power) a predictive biomarker for two possible biomarker measurements errors, several QR‐based responder definitions and QR as a continuous variable. The QR‐based responder definitions were determined by the thresholds from the percentiles of the normal distribution (in 5% increments) where the mean was set to the QR control group mean and the variation (variance) was the combined variation from both treatment groups (pooled variance).
Statistical power (via simulation) to detect a predictive biomarker by several responder definitions (and QR) and two measurement errors of the biomarker. The treatment effect is proportional to the biomarker (minimum of no effect and a maximum of 2.36Δ)
The continuous QR produced the highest chance of detecting a predictive biomarker (statistical power) regardless of the variation in measuring the biomarker (measurement error). This was true with other levels of variation and when the biomarker distribution was not bell‐shaped (see Appendix S1 : Table S1 ). The chance of detecting a predictive biomarker (statistical power) when dichotomizing the QR for the various percentile‐based responder definitions varied considerably. From Table Table2, 2 , the highest chance of detecting a predictive biomarker occurs at the 70th percentile definition of responder regardless of the precision of measuring the biomarker (measurement error). It also remains the maximum chance of detecting (greatest statistical power) when the biomarker distribution is other than bell‐shaped (see Appendix S1 : Table S1 ). Nonetheless, the continuous QR provides the maximum chance of detecting a predictive biomarker over any of these responder definitions. When relaxing the level of significance from 0.05 to 0.10 (offered as a solution when testing interactions and statistical power is low), the chance of detecting a predictive biomarker when analysed as a continuous variable comes close to conventional levels (0.720 and 0.746 when the measurement error of the biomarker is σ and σ /2, respectively. See Appendix S1 : Table S2 ).
We have shown that the analysis of covariance (ANCOVA) model of 1‐year log‐transformed, age‐adjusted, C‐peptide is consistently good predictor across several TrialNet studies. We defined the quantitative response (QR) as the observed 1‐year C‐peptide minus the model‐based expected C‐peptide level. We confirmed the excellent behaviour of QR using a few of the studies used in fitting the model as well as a trial that was independent of the modelling. 12 Defined in this way, a positive shift in the QR distribution provides the magnitude of the treatment effect on C‐peptide for an ‘active’ treatment, while the QR mean is around zero for the placebo group (or an inactive treatment). The QR is calculated at the subject level and so provides a visually informative way of viewing all subjects in the study and the treatment effect. In addition, the QR allows for a simple analytical test of treatment effect consistent with the standard ANCOVA model test.
Without any a priori biological basis, there is little justification for choosing any particular threshold to partition the treated group into distinct categories of responder and non‐responder. This is particularly true if the interpretation is to identify subjects that had treatment benefit from those that did not (or benefited minimally). We offer several arguments to support this contention. One, some of the individuals classified as ‘responders’ may have attained their QR value (or less C‐peptide decline) because they have inherently less severe disease (without considering the benefit they received from treatment). Two, when setting the threshold at a very stringent level (eg no C‐peptide decline), those classified as ‘responders’ have a higher likelihood of having exaggerated levels of C‐peptide. This is a proven statistical phenomenon referred to as regression to the mean. Three, it is possible that subjects that were destined to have low C‐peptide levels if not treated, had substantial benefit due to treatment but still not greater than the selected threshold used to define responder. Thus, the degree of misclassification due to dichotomizing may be substantial and misclassification may go either way. Our simulation studies clearly indicate that using the QR in its continuous form will increase the chance of discovering a biomarker correlated with treatment effect, that is, a predictive biomarker.
Nonetheless, if a compelling reason remains to group subjects as responders and non‐responders, then using the placebo group's 70th or 65th percentile of the QR distribution as a threshold to partition the experimental treatment group into responder categories provides the smallest reduction in the chance of detecting a predictive biomarker. We suggest using the placebo or control group to define the threshold and did so in our simulations. Not presented were other ways of determining a QR threshold to define response. The power was slightly less (2 to 3%) than the values displayed in Table 2 when using the empirical percentiles of QR from the placebo group. Alternatively, using a fixed percentile threshold taken from the normal distribution with mean of zero and standard deviation of 0.151 (determined from the modelled cohort) provided a slightly higher statistical power than in Table 2 but only by less than 1%. However, this fixed threshold ignores the QR distribution from the placebo group of the trial analysed. Determining thresholds using the QR distribution from the internal control group adjusts for any possible shift that may occur in a future trial.
It is imperative that investigators involved in the analysis of biomarkers in the context of a clinical trial understand the difference between a prognostic and a predictive biomarker. It is essential that any biomarker that is correlated with QR in the treated group be evaluated in the placebo group. We suggest to model the QR with two covariates, treatment group and the biomarker to be evaluated, as well as an interaction term; this allows a way to quantify the predictive from the prognostic effect of the biomarker. The interpretation of the biomarker's utility will be dramatically different. If a biomarker is prognostic, it will be advisable to measure this marker in subsequent trials in order to adjust for it in the analysis. If the biomarker is predictive, it will likely have value for targeting subjects for further study, particularly in primary prevention trials of the agent associated with the biomarker.
Our analytical approach provides less statistical power (see Table Table2) 2 ) than approaches that ignore the distinction between predictive and prognostic biomarkers. Testing for an interaction effect term in any model is always subject to less power than the main effect terms. Testing at a relaxed level of significance (α = 0.10) increases the statistical power to an acceptable level. While rectifying the low power, such an adjustment increases the chance of a false‐positive result. In addition, these studies usually evaluate multiple biomarkers (multiple testing problem) that further contributes to the risk of one or more false positives. In dealing with both circumstances, a reasonable strategy would be not to adjust for multiple tests but rather consider the detection of any predictive biomarker as hypothesis generating and pursue confirmation in an independent setting. One weakness of the QR is that there remains a fair amount of unexplained variance. We have suggested that using our method may identify a biomarker as predictive. Alternatively, our approach could identify a biomarker as prognostic and therefore lead to a revised QR equation which would have reduced unexplained variance, that is, improved prediction at the subject level. The QR units are not meaningful to investigators except that a positive value represents a subject that had higher 1‐year c‐peptide level than what was expected from their age and baseline c‐peptide, and conversely, a negative QR means lower 1‐year c‐peptide than expected. One caution is in order, if an agent provides differential effect across age, then using the QR method may not be prudent at least until one quantifies its predictive effect. For this reason, we suggest checking age as a predictive variable using the observed change in c‐peptide and then proceed using age‐adjusted QR method if there is no convincing evidence of such an effect.
Future plans are to apply the QR method to empirical data in which biomarkers have been measured from the samples in both treated and controls. Furthermore, to address the timing of when the biomarker was measured (baseline and after/during treatment), this timing not only changes the interpretation of the results but also changes the analytical methods employed.
CONFLICT OF INTEREST
The authors declare that they have no conflicts of interest. Drs. Bundy and Krischer work at the TrialNet Coordinating Center funded by the NIDDK of the NIH.
Brian Bundy and JP Krischer researched data, contributed to discussion, wrote the manuscript and reviewed/edited the manuscript.
All TrialNet protocols included in this study were approved by each participating institution's Institutional Review Board (IRB) or Ethics Committee/Research Ethics Board (EC/REB). The NIH (National Institute of Diabetes and Digestive and Kidney Diseases) was the sponsor for these trials.
Brian Bundy and JP Krischer are the guarantors of this work, had full access to all the data and take full responsibility for the integrity of data and the accuracy of data analysis.
The sponsor of the trial was the Type 1 Diabetes TrialNet Study Group. The Type 1 Diabetes TrialNet Study Group is a clinical trials network funded by the National Institutes of Health (NIH) through the National Institute of Diabetes and Digestive and Kidney Diseases, the National Institute of Allergy and Infectious Diseases and The Eunice Kennedy Shriver National Institute of Child Health and Human Development, through the cooperative agreements U01 DK061010, U01 DK061034, U01 DK061042, U01 DK061058, U01 DK085465, U01 DK085453, U01 DK085461, U01 DK085466, U01 DK085499, U01 DK085504, U01 DK085509, U01 DK103180, U01 DK103153, U01 DK085476, U01 DK103266, U01 DK103282, U01 DK106984, U01 DK106994, U01 DK107013, U01 DK107014, UC4 DK106993, UC4 DK11700901 and the JDRF. The contents of this article are solely the responsibility of the authors and do not necessarily represent the official views of the NIH or the JDRF.
Bundy BN, Krischer JP; for the Type 1 Diabetes TrialNet Study Group . A quantitative measure of treatment response in recent‐onset type 1 diabetes . Endocrinol Diab Metab . 2020; 3 :e00143 10.1002/edm2.143 [ CrossRef ] [ Google Scholar ]
DATA AVAILABILITY STATEMENT
- Open access
- Published: 01 August 2022
Quantitative and qualitative analysis of the quality of life of Type 1 diabetes patients using insulin pumps and of those receiving multiple daily insulin injections
- Lilian Tzivian 1 ,
- Jelizaveta Sokolovska 1 ,
- Anna E. Grike 2 ,
- Agate Kalcenaua 3 , 4 ,
- Abraham Seidmann 5 , 6 ,
- Arriel Benis 7 , 8 ,
- Martins Mednis 4 ,
- Ieva Danovska 4 ,
- Ugis Berzins 4 ,
- Arnolds Bogdanovs 4 &
- Emil Syundyukov 4 , 9
Health and Quality of Life Outcomes volume 20 , Article number: 120 ( 2022 ) Cite this article
Insulin pump therapy represents an alternative to multiple daily injections and can improve glycemic control and quality of life (QoL) in Type 1 diabetes mellitus (T1DM) patients. We aimed to explore the differences and factors related to the T1DM-specific QoL of such patients in Latvia.
Design and methods
A mixed-method cross-sectional study on 87 adult T1DM patients included 20 pump users and 67 users of injections who participated in the quantitative part of the study; 8 pump users and 13 injection users participated in the qualitative part. Patients were invited to participate using a dedicated digital platform. Their QoL and self-management habits were assessed using specially developed questionnaires adapted to Latvian conditions. Multiple logistic regression models were built to investigate the association between social and self-management factors and patients’ QoL. In addition, qualitative analysis of answers was performed.
Insulin pump users were younger, had higher incomes, and reported higher T1DM expenses than users of multiple daily injections. There were no differences in self-management between the groups; Total QoL differed at the 0.1 significance level. In fully adjusted multiple logistic regression models, the most important factor that increased Total QoL was lower T1DM-related expenses (odds ratio, OR 7.02 [95% confidence interval 1.29; 38.0]). Men and those with more years of living with T1DM had better QoL (OR 9.62 [2.20; 42.1] and OR 1.16 [1.05; 1.29], respectively), but the method of administration was not significantly associated with QoL (OR 7.38 [0.87; 62.9]). Qualitative data supported the results of quantitative analysis.
QoL was the main reason to use an insulin pump, while the expense was the main reason to avoid the use of it or to stop using it. Reimbursement policies thus should be considered to enable patients to choose the more convenient method for themselves.
Type 1 diabetes mellitus (T1DM) is a chronic autoimmune characterized by hyperglycemia due to loss of insulin producing cells of the pancreas that can end in diabetic coma and eventually death [ 1 , 2 ]. T1DM incidence has increased on average 3–4% over the past 30 years [ 3 ], reaching an incidence of 15 people per 100,000 and a prevalence of 9.5 per 10,000 worldwide [ 4 ]. In Latvia, there were 4169 patients with T1DM in 2015 (prevalence of 211.7 per 100,000), and an incidence of 13.5 people per 100,000 [ 5 ].
The main therapy for T1DM patients is insulin regulation via multiple daily injections or continuous subcutaneous infusions using an insulin pump. Patients aim for glycated hemoglobin (HbA1c) levels below 7% [ 1 ] without an unacceptable incidence of hypoglycemia [ 6 ]. This process demands a certain amount of self-management, such as treatment diaries and recording and interpretation of blood sugar levels. Some patients, however, struggle with these tasks and fail to successfully continue a therapy, especially in the case of multiple daily injections. The use of an insulin pump as a technological solution can simplify efforts to manage the process and to maintain desired levels of blood glucose [ 7 ].
Administration of insulin via a pump improves glycemic control with fewer hypoglycemic episodes in T1DM subjects previously conventionally treated with multiple daily injections, achieving a significant reduction in HbA1c. Meta-analyses reveal that in patients treated with an insulin pump, Hb1A1c decreased more pronouncedly and reported insulin requirements were lower [ 8 ] than for injection patients, especially young children. Severe hypoglycemia episodes were rare, indicating better glycemic control and lower incidence of nocturnal hypoglycemia [ 8 , 9 , 10 ].
The quality of life (QoL) of patients with T1DM is affected by complications and fear of them and is lower than that of healthy peers [ 11 ]. Using an insulin pump reduces fear of severe hyperglycemia and diabetic coma [ 12 , 13 ]. Patients using a pump have more flexible possibilities regarding meals, diet, everyday activities, and socialization [ 14 ], as the pump supports improved self-management habits [ 15 , 16 , 17 ]. Some additional non-health-related benefits, such as reduced worry about supplies while traveling, can significantly improve patients’ QoL as well [ 12 ]. However, the pump itself and related physical restrictions can be mentioned as disadvantages [ 11 , 14 ]. The most prominent problem with an insulin pump is the expense. There is a large difference in cost between injections and a pump. Although studies show that there is a good value for money in the use of a pump, many adult patients may be unable to afford one [ 12 ].
According to the Latvian Diabetes Association, 3700 patients in Latvia currently have T1DM [ 18 ]. Insulin pumps are covered by the state until 18 years old [ 19 ], but adult patients must pay for the pump itself (approximatively 3500 EUR) and also cover the cost of pump-related disposables, amounting to more than 100 EUR per month. Considering the low-income level in Latvia (an average of 583 EUR per household member per month, in 2020) [ 20 ], insulin pump therapy is a huge financial burden. In these circumstances, the investigation of factors related to the QoL of patients using different methods of insulin administration can identify appropriate changes to reimbursement policies to improve such patients’ disease-related conditions.
The aim of this mixed-method cross-sectional study was to compare the QoL and T1DM-related self-management of two groups of patients residing in Latvia—insulin pump users and those who use multiple daily insulin injections, and to investigate factors associated with their QoL. Our main hypotheses were as follows:
The QoL of insulin pump users is better than that of injection users, and T1DM-related self-management is easier for pump users than injection users.
Easier T1DM-related self-management is associated with better QoL.
We investigated also specific reasons to use or not the pump or injections, including the reasons for changes between different methods of insulin administration, using both qualitative and quantitative methodologies. Our main hypothesis was that the major reason for using a specific method of administration and for changes in the method used is treatment-related expenses.
Research design and methods
Study design and population.
The mixed-method cross-sectional study was conducted in April and May 2021 and consisted of a quantitative part and a qualitative part. We chose a combined approach due to the small number of insulin users in Latvia that lead to imprecision in the quantitative results. All T1DM patients at least 18 years of age who signed informed consent forms were eligible to participate in the study. As the total number of adult insulin pump users in Latvia is very small (about 40 users to our knowledge), we invited all of them to participate in the quantitative part of this study. The number of multiple injection users was planned to be in a proportion of 1:2 according to the enrolled sample of insulin pump users, and the calculated power of the study in that case was 80%. For the qualitative part of the study, the number of participants depended on their agreement and on the saturation of interviews—a lack of new information collected during the additional interviews. The saturation was defined by the investigator during the interviewing process. Both groups of participants—those using insulin pumps and those receiving multiple insulin injections—were enrolled in the qualitative part of the study. The study was approved by the Scientific Research Ethic Commission of the Institute of Cardiology and Regenerative Medicine of the University of Latvia on February 2, 2021.
Methods of enrolment of the study participants
Patients learned about the study from e-mail materials received from doctors, patient organizations’ representatives, or diabetes nurses. Additionally, they could learn about the study from posts in the closed Facebook group “Diabetes in Latvia”. We also identified potential study participants using metadata from the longitudinal study “LatDiane: Latvian diabetic nephropathy study”, initiated in 2013 [ 21 ]. Currently, more than 355 well-characterized patients with Type 1 diabetes are in the LatDiane study. Invitations included a description of the study and its objectives, as well as a technical guide for onboarding on the digital platform developed for this study [ 22 ]. Patients were invited to participate using a digital engagement platform, equipped with a dynamic e-consent management tool (Fig. 1 ). The web-based and mobile-ready engagement platform was developed as a collaboration among clinicians, epidemiologists, and data protection and digital health specialists. The website of this study includes detailed instructions in Latvian and Russian, conditions for participation, information regarding the aims and organization of the study, and a contact section. Participants were asked to provide their consent (which could be dynamically managed on the platform, e.g., for opt-out) for data processing, in compliance with the General Data Protection Regulation (GDPR) [ 23 ].
Visual engagement material used in study invitations
Once they provided their informed consent, participants were invited to complete the online questionnaire and were informed about the time slots available for semi-structured interviews. After data collection, the system extrapolated a dataset that described the user survey input results, fully separated from the actual database (Fig. 2 ).
Electronic platform—research study metadata query view
The quantitative part of the study
The quantitative part of the study included self-reported socio-demographic information (age, gender, education, living conditions, financial needs, and income) and disease-related factors, including weight and height for calculation of the body-mass index (BMI) (Additional file 1 : Supplement 1), years of living with T1DM, number of hypoglycemia incidents per week, number of hypoglycemia incidents per half-year, HbA1c in the last medical check, number of HbA1c checks during the year, and T1DM expenses (Additional file 1 : Supplement 2).
We developed questionnaires for this study that consider conditions in Latvia. The QoL questionnaire comprised 35 questions divided into five blocks: Signs and symptoms (15 questions), Therapy (5 questions), Care (6 questions), Concerns (4 questions), and Communication (6 questions) (Additional file 1 : Supplement 3). The self-management questionnaire consisted of 19 questions divided into three blocks: General, Diet, Physical activities (Additional file 1 : Supplement 4).
All questionnaires were available in the two main languages in use in Latvia – Latvian and Russian. Translation and back translation of questionnaires were performed by two independent professional translators.
Statistical analysis of the quantitative part
The reliability of the questionnaires was checked using the alpha-Cronbach’s test (α) after the first 20 participants had responded (α > 0.75 for all blocks of the questionnaire). These participants were subsequently included in the study sample, and their answers were analyzed together with those of other participants. For both questionnaires, we transformed the answers into values between 0 and 100 and then calculated means for each block. We further calculated the Total QoL and Total SM (self-management) scores as the means of all questions in their respective surveys. Higher values mean better QoL or better self-management.
We next compared pump users and injection users for all demographic variables, using central and dispersion measures according to the type of each variable. We used the Mann–Whitney test to compare qualitative variables and Chi-squared or Cramer’s V tests to compare the quantitative ones. We investigate the correlation between individual subscales and Total QoL and Total SM using Spearman correlation. We considered a two-sided 0.1 significance level for this stage of analysis.
We built multiple logistic regression models for Total QoL, dividing the Total QoL variable at the median (‘worse’ ≤ 67.9’, ‘better’ > 67.9). Variables found univariately statistically significantly related to Total QoL at the 0.1 significance level were included in logistic regression models together with demographic and T1DM-related variables that were found significantly different between pump users and injection users. The full adjustment set included the method of administration, age, sex, education, income, T1DM expenses, years with T1DM, and Total SM. We choose the best model fit according to the − 2 Log-likelihood test. p value < 0.05 was considered statistically significant for this part of the study. Odds ratio (ORs) and 95% confidence intervals were presented for multiple logistic regression models. We used Statistical Package for Social Science (SPSS) software (26th version) for the statistical analysis [ 24 ].
Additional and sensitivity analyses
For additional insight, we asked pump users about the number of years they have used the pump and their reasons for using one (6 categories: QoL, insulin dosing, less pain, less hypoglycemia, just trying, and other). We asked injection users about reasons for not using a pump (6 categories: no trust, expensive, lack of appropriate model, lack of willingness, negative information, other) and reasons for ceasing to use a pump if it was used previously (5 categories: expensive, lack of trust, not comfortable, not resultative, and other). For the sensitivity analysis, we built multiple logistic regression models for two QoL blocks that significantly differed between the user groups (Therapy and Concerns), dividing the results for each block by the median value into ‘worse’ and ‘better’ and using the same set of covariates.
Qualitative part of the study
The qualitative part of the study consisted of analysis of semi-structured interviews performed face-to-face or via telephone or video chats. Interviews were recorded, coded, transcribed according to their major theme, and analyzed using Nvivo software (version 12) [ 25 ] to obtain subcategories of each major theme. Coding of interviews included changing participants’ names. In this paper, we provide part of the results of the qualitative analysis as support for interpreting the quantitative results.
We enrolled 87 T1DM patients in the quantitative part of the study: 20 pump users and 67 injection users. Both groups included mostly women. Pump users generally had at least some secondary education and had higher incomes, while injection users mostly had just a high school education. Pump users were younger (mean age 21.5 years, standard deviation (SD) 4.4) than injection users (mean age 33.6 years, SD 11.0). The groups did not differ by other socio-demographic characteristics.
HbA1c values at the last medical check did not differ significantly between the groups. In both groups, most of the patients performed one medical check during 2020 and till May 2021. Only 10% of pump users and 15% of injection users mentioned four medical checks during this period. There were no differences between pump users and injection users in this parameter. T1DM-related expenses were statistically significantly higher for pump users: for 94.7% of them, these expenses were more than 100 EUR/month; just 29.2% of injection users had similarly high expenses (Table 1 ).
Quality of life and self-management
The reliability of all scales was high, ranging from α = 0.75 to α = 94 for all blocks for both questionnaires (excluding the SM Diet that had medial reliability; α = 0.63). Correlation between QoL and self-management was weak and partly insignificant. Self-management blocks correlated among themselves significantly, but not strongly (Additional file 1 : Table S1). No significant relations were found between the number of tests and three self-management blocks ( p = 0.12, p = 0.86, and p = 0.36, respectively).
Significant differences at the 0.1 significance level were observed between user groups in their Therapy and Concerns blocks, and in Total QoL. The highest values for both groups were found for Therapy and Communication blocks of QoL. There were no significant differences between groups in their self-management blocks (Table 2 ). Univariate relationships were found between Total QoL and sex ( p = 0.03).
In fully adjusted multiple regression models, pump users were seven times more likely to have a high Total QoL than injection users (OR 7.38; CI 0.87; 62.9). Factors that increased Total QoL were lower age, male sex, lower T1DM expenses (the most prominent association), more years living with T1DM, and better self-management. Most of the confidence intervals were wide, pointing to the low number of participants in the study (Table 3 ). However, the post hoc calculated power of analysis was 70.1% ( p = 0.01), indicating the study’s medial power.
For pump users, the main reason to use a pump was improved QoL; this was mentioned by 90% of them. For injection users, the median time they had been using insulin injections was eight years, and the main reason for not using a pump was its cost, as mentioned by almost half of these respondents. Of the 13 patients that previously used a pump, the main reason why they stopped was the cost (mentioned by 46.2% of those that stopped using a pump) (Additional file 1 : Table S2).
In the univariate analysis between the Therapy block of QoL and demographic and T1DM-related factors, significant relationships at the 0.1 significance level were found for years with T1DM ( p < 0.01) and T1DM expenses ( p = 0.08); for the Communication block, significant univariate relationships were found for the number of hypoglycemic episodes per week ( p = 0.09) and sex ( p = 0.07). Consistent with the main analysis, male sex, lower T1DM expenses, and years living with T1DM were associated with better Therapy and Communication blocks (Additional file 1 : Table S3).
Of those included in the quantitative part of the study, 8 pump users and 13 injection users also participated in the qualitative interviews; 15 of these were women. The men-women proportion in each study arm was similar to that in the quantitative part of the study.
The age of the interviewees ranged from 18 to 50 years, and years with T1DM ranged from 1 to 35. Eight participants did not have T1DM diaries, three had one only at the beginning of their treatment, two use them only for visits with a physician, and six regularly recode their activities in their diaries (two using an app to do so). One participant kept a diary when she used multiple insulin injections but stopped when she switched to an insulin pump (Table 4 ).
Analysis of 40 identified codes of the interviews revealed three major themes of answers: diagnosis-related, daily self-management, and life with T1DM. Each of the major themes was further divided into three to four subcategories (Table 5 ). Here we will present a part of the results related to one subcategory for each category of answers: perception of diagnosis (major theme: diagnosis-related), insulin administration (major theme: daily self-control), and T1DM-related costs (major theme: life with T1DM).
Perception of diagnosis
Before their diagnosis, most participants had had some symptoms that they had not related to T1DM, such as thirst, frequent urination, weight loss, and weakness. Therefore, for nearly all of them, the diagnosis was unexpected and shocking. For example, I, who was diagnosed at the age of 28 after being hospitalized due to T1DM:
I didn't know anything before, it seemed to me that diabetes could be born or not. I was so bad in that resuscitation because I was in a severe hypoglycemic condition … my head was dull … it was so hard to grasp.
This reaction was not related to the participant’s age at the time of diagnosis (Additional file 1 : Supplement 5).
One of the main reasons to use an insulin pump was the QoL that it provides (Additional file 1 : Supplement 6). For example, M said:
I have much more control with the pump, because I can adjust insulin doses if necessary, and adjust the time for basal insulin. I can stop insulin if needed. with the syringe, you are injecting and then you can no longer control what is going. The pump gives much more control to both the doctor and the patient, if a person understands how the pump works. But that's what training is for.
However, some of the injection users saw positive aspects in their treatment method as well. For example, I, who uses the injections:
When using injections, it is nice to inject insulin once and that is .
F, who has used injections for 31 years, was categorically opposed to the idea of a pump:
No, never! It is not practical for me to have a foreign object that is always present at my waist area. I feel very uncomfortable. That limits me .
To summarize: although QoL was mentioned by most of the participants as the determining factor for use of the pump, some participants feel that a pump is less comfortable and even disturbing. This supports the quantitative result showing a lack of proper relations between the method of administration and QoL.
Most pump users in our study mentioned the cost of this administration method (Additional file 1 : Supplement 7). For some participants, the decision whether to use a pump depends on the monthly costs. For example, K said:
It is an extra investment [talking about the pump] —now I have needles and insulin for free, I do not have to buy anything extra—just those test strips, because the glucometer is also free for me. Together it's pretty affordable .
D switched from the pump to injections several times because of financial problems:
I had already used it [pump] as a child, I was 13 years old. […] I used to have insulin pens, but then my mom saved money so I could have the pump. […] After that I had to switch back to insulin pens because I was in big financial trouble. However, I really wanted to get back to the pump.
In Latvia, state reimbursement for insulin pumps is possible until the age of 18. Thus, some people are forced to switch to injections at that point. For most of the participants who would like to use an insulin pump, treatment-related costs are too high, and some of them were forced to change to the cheaper injection method. This supports the quantitative result of the study on the relation between T1DM-related costs and QoL.
Discussion and conclusions
In this study, we investigated quantitatively and qualitatively factors related to the QoL of patients with T1DM according to their method of insulin administration: using an insulin pump or using multiple daily injections. The reported QoL was found to be associated with the method of insulin administration, the age and sex of the participants, the number of years the patient had lived with T1DM, self-management, and T1DM-related expenses. QoL was the main reason cited for using a pump, while the expense was the main reason to avoid its use or to stop using it.
An association between the method of insulin administration and the QoL of patients with T1DM has been shown previously both in qualitative and in quantitative studies [ 11 ]. However, until recently, most of the studies on insulin pumps were qualitative and were performed on populations of children [ 26 , 27 , 28 ]. In the last decade, quantitative evaluations of pump use had appeared as well, but studies combining these two methods of investigation are still scarce. However, similarity among their objectives allows us to combine the results of different studies to provide additional explanations of our observed results. For example, Alqambar et al. found higher scores for QoL for pump users than for injection users. The former had significantly higher satisfaction with their treatment and had a lower burden of disease (both with p < 0.01) [ 29 ]. These results are supported by the qualitative study by Mesbah et al., which described higher satisfaction among pump users in many areas [ 22 ]. In our study, although we did not observe statistically significant differences in QoL between pump and injections users, QoL was the main reason given for using the pump. Nevertheless, in our study some participants had a negative attitude toward the pump. Mesbah et al. likewise report the existence of negative feelings toward pumps, such as fear of being dependent on a machine or concern about sporadic mechanical problems [ 30 ].
As QoL is multidimensional, factors affecting it might differ according to study design and measures. For example, in our study we did not observe any association of QoL with the level of HbA1c. In contrast, in the study by Alavrado-Martel et al. performed in Spain, worse QoL was associated with increasing HbA1c [ 31 ]. This fact is extremely interesting, as in both studies the mean age of participants was 31 years and mean years living with T1DM were 14, and participants had similar levels of education. It is possible that the difference can in part be explained by the proportion of pump users: a third of our participants use a pump and therefore are in reduced risk of an increased level of blood sugar, but in the Spanish study only 5% of patients were pump users. Therefore, the association with the level of HbA1c was not prominent. In addition, in the Spanish study a better QoL was associated with the female sex, but in our study, it was associated with the male sex. As mentioned by Mesbah et al., lack of flexibility in clothing options can reduce the QoL of female pump users [ 30 ], and this may be reflected in our results. In another study [ 32 ] women with diabetes were found to evaluate their health status and diabetes-related care worse than men; they also had more diabetes-related worries related to higher levels of Hb1Ac, although their level of metabolic control did not differ from that of men.
Initiation of pump therapy in Latvia usually is not a choice, but a costly necessity due to problems in diabetes management (such as hypoglycemia) and discomfort associated with diabetes treatment (e.g., pain, fear of injections). Studies describe substantial clinical benefits of insulin pumps for such patients. For example, in the meta-analysis by Benkhadra et al. based on 25 randomized clinical trials, absolute HbA1c reduction was better managed in pump users than in injection users (difference of 37%, CI 0.24; 0.51), and this result was consistent across adults and children. In addition, pump users had a lower risk of hypoglycemia (relative risk, RR = 0.85, CI 0.6–1.2) [ 10 ]. These results were supported by another meta-analysis by Jeitler et al. that analyzed 33 studies and found a 43% difference (CI – 0.65; – 0.20) between groups with different methods of insulin administration [ 8 ]. In addition to clinical benefits, the patient’s ability to self-manage should be considered when choosing the method of administration. Our study did not observe any difference in self-management between pump users and injection users, but we do observe a slight but significant increase in QoL for those with better self-management. The main cause seems to be educational training provided for all patients with T1DM. Previous studies have described the effectiveness of such training on self-management. For example, in the structural analysis by Campbell et al. based on 18 studies, people who attended educational training gained clinical benefits by managing their lives according to the knowledge they received during these sessions. However, people were often tired and encountered difficulties in managing their everyday lives according to guidelines even during these educational trainings, which made additional follow-up by the physician essential [ 7 ]. For pump users less intensive follow-up is needed, thus removing a level of stress from both physician and patient. Overall, a personal approach when choosing the method of insulin administration seems to be the best in the case of T1DM patients.
Limitations of the study
The main limitation of our study is its cross-sectional nature, which does not allow us to evaluate causal relationships. The study included a small number of participants, especially in the pump-users group, and we did not divide participants into age groups. Further, the use of the Internet for enrollment limited the available pool of participants and may introduce a volunteer bias that can affect the validity of the results. Further limitations include self-evaluation of QoL and self-management and possible errors regarding the number of medical checks due to memory bias. Specific questions on installation, operation, troubleshooting, and handling of the insulin pumps (these factors could affect the quality of therapy in pump-users and have an impact on QoL) were not included in the study to avoid complexity. In addition, some limitations in the qualitative part of the study could be related to the language, as the native language of the interviewer was Latvian. Despite the good knowledge of Russian, some impreciseness could occur.
Although the reliability of all parts of the survey was high at the initial stage of their check, we observed the medial reliability of one of its parts after collecting the information about all study participants. As we did not see any difference between the insulin pump users and multiple injection users in other parts of the SM questionnaire, we assumed that the lower reliability of this part of the questionnaire will not affect the results of our study.
Another limitation of our study is the high proportion of insulin pump users. Before we start the study, we knew that the number of insulin pump users in the Latvian population is relatively small. Therefore, we decided to invite participants in the proportion of 1:3 (pump users versus injection users) to increase the overall power of analysis. We attempt to invite as many pump users as it was feasible. As a result of this strategy, we observed a disproportion between the pump users and injection users in their relation to the whole pump/injection users’ population in the country. This can affect the results of our study, especially the qualitative part of them.
Strength of the study
A major strength of our study is a mixed methodology that allows us to describe QoL-related parameters of T1DM patients from various sides. Further, although the sample size was small, it represented more than half of all pump users in Latvia, and regardless of the small sample size, the power of analysis was 70%.
QoL was the main reason to use an insulin pump, while the main reasons to avoid one were expenses related to its use. However, the expense related to diabetes treatment, not the method of insulin administration, was the strongest predictor of T1DM patients’ QoL. Reimbursement policies thus should not only consider the patient’s personal preference for treatment, but also be structured to alleviate ongoing maintenance costs, particularly as high costs drive reduced adherence to treatment regimens that in turn impose higher costs on the healthcare system in the form of additional disorders and comorbidities.
Development of national insurance policies is critical worldwide, but especially in countries like Latvia with overall weak health care and public health systems, supporting reimbursement for insulin pumps could help:
to reduce complications related to poor treatment adherence,
to avoid increased additional morbidity, and
to prevent an overload of the health system.
Availability of data and materials
The datasets generated and/or analyzed during the current study are available in the Longenesis Curator platform, accessible via following link: www.longenesis.com/curator .
Type I diabetes mellitus
- Quality of life
95% Confidence interval
Janež A, Guja C, Mitrakou A, et al. Insulin therapy in adults with type 1 diabetes mellitus: a narrative review. Diabetes Ther. 2020;11(2):387–409. https://doi.org/10.1007/s13300-019-00743-7 .
Article PubMed PubMed Central Google Scholar
Katsarou A, Gudbjörnsdottir S, Rawshani A, et al. Type 1 diabetes mellitus. Nat Rev Dis Primers. 2017;3(1):17016. https://doi.org/10.1038/nrdp.2017.16 .
Article PubMed Google Scholar
Norris JM, Johnson RK, Stene LC. Type 1 diabetes—early life origins and changing epidemiology. Lancet Diabetes Endocrinol. 2020;8(3):226–38. https://doi.org/10.1016/S2213-8587(19)30412-7 .
Article CAS PubMed PubMed Central Google Scholar
Mobasseri M, Shirmohammadi M, Amiri T, Vahed N, Hosseini Fard H, Ghojazadeh M. Prevalence and incidence of type 1 diabetes in the world: a systematic review and meta-analysis. Health Promot Perspect. 2020;10(2):98–115. https://doi.org/10.34172/hpp.2020.18 .
Centre of Disease Prevention and Control https://www.spkc.gov.lv/lv/latvijas-veselibas-aprupes-statistikas-gadagramata-2015/3_sabiedribas_veseliba_2015_11.pdf (Last visited: 26 December 2021)
Pickup JC. Is insulin pump therapy effective in type 1 diabetes? Diabet Med. 2019;36(3):269–78. https://doi.org/10.1111/dme.13793 .
Article CAS PubMed Google Scholar
Campbell F, Lawton J, Rankin D, et al. Follow-Up Support for Effective type 1 Diabetes self-management (The FUSED Model): A systematic review and meta-ethnography of the barriers, facilitators and recommendations for sustaining self-management skills after attending a structured education programme. BMC Health Serv Res. 2018;18(1):898. https://doi.org/10.1186/s12913-018-3655-z .
Jeitler K, Horvath K, Berghold A, et al. Continuous subcutaneous insulin infusion versus multiple daily insulin injections in patients with diabetes mellitus: systematic review and meta-analysis. Diabetologia. 2008;51(6):941–51. https://doi.org/10.1007/s00125-008-0974-3 .
Retnakaran R, Hochman J, DeVries JH, et al. Continuous subcutaneous insulin infusion versus multiple daily injections: the impact of baseline A1c. Diabetes Care. 2004;27(11):2590–6. https://doi.org/10.2337/diacare.27.11.2590 .
Benkhadra K, Alahdab F, Tamhane SU, McCoy RG, Prokop LJ, Murad MH. Continuous subcutaneous insulin infusion versus multiple daily injections in individuals with type 1 diabetes: a systematic review and meta-analysis. Endocrine. 2017;55(1):77–84. https://doi.org/10.1007/s12020-016-1039-x .
Pouwer F, Hermanns N. Insulin therapy and quality of life. A review Diabetes Metab Res Rev. 2009;25(S1):S4–10. https://doi.org/10.1002/dmrr.981 .
Cummins E, Royle P, Snaith A, et al. Clinical effectiveness and cost-effectiveness of continuous subcutaneous insulin infusion for diabetes: systematic review and economic evaluation. Health Technol Assess. 2010. https://doi.org/10.3310/hta14110 .
Rechenberg K, Whittemore R, Grey M. Anxiety in youth with type 1 diabetes. J Pediatr Nurs. 2017;32:64–71. https://doi.org/10.1016/j.pedn.2016.08.007 .
Alsaleh FM, Smith FJ, Taylor KM. Experiences of children/young people and their parents, using insulin pump therapy for the management of type 1 diabetes: qualitative review: experiences of using insulin pumps. J Clin Pharm Therap. 2012;37(2):140–7. https://doi.org/10.1111/j.1365-2710.2011.01283.x .
Article CAS Google Scholar
Abualula NA, Jacobsen KH, Milligan RA, Rodan MF, Conn VS. Evaluating diabetes educational interventions with a skill development component in adolescents with type 1 diabetes: a systematic review focusing on quality of life. Diabetes Educ. 2016;42(5):515–28. https://doi.org/10.1177/0145721716658356 .
Ross LJ, Neville KA. Continuous subcutaneous insulin infusion versus multiple daily injections for type 1 diabetes. J Paediatr Child Health. 2019;55(6):718–22. https://doi.org/10.1111/jpc.14480 .
Thabit H, Hovorka R. Continuous subcutaneous insulin infusion therapy and multiple daily insulin injections in type 1 diabetes mellitus: a comparative overview and future horizons. Expert Opin Drug Deliv. 2016;13(3):389–400. https://doi.org/10.1517/17425247.2016.1115013 .
http://diabetam.lv/2018/06/07/skaidrojums-par-kriterijiem-insulina-pumpja-terapijas-piemerosanai/ (Last visited: 26 December 2021)
Central Statistical Bureau (CSB) - https://lvportals.lv/dienaskartiba/323225-2019-gada-majsaimniecibu-riciba-esosie-ienakumi-pieauga-par-68-2020 (Last visited: 26 December 2021)
Salna I, Salna E, Pahirko L, Skrebinska S, Krikova R, Folkmane I, Pīrāgs V, Sokolovska JJ. Achievement of treatment targets predicts progression of vascular complications in type 1 diabetes. Diabetes Compl. 2021;35(12):108072.
Article Google Scholar
Digital Engagement Platforms, Longenesis Curator. Available: ttps:// www.longenesis.com/curator ; Longenesis Engage. Available: https://www.longenesis.com/engage
Regulation (EU) 2016/679 of the European Parliament and of the Council of 27 April 2016 on the protection of natural persons with regard to the processing of personal data and on the free movement of such data, and repealing Directive 95/46/EC (General Data Protection Regulation), OJ 2016 L 119/1.
IBM Corp. Released 2019. IBM SPSS Statistics for Windows, Version 26.0. Armonk, NY: IBM Corp.
QSR International Pty Ltd. 2018; NVivo (Version 12), https://www.qsrinternational.com/nvivo-qualitative-data-analysis-software/home
Valenzuela JM, Patino AM, McCullough J, et al. Insulin pump therapy and health-related quality of life in children and adolescents with type 1 diabetes. J Pediatric Psychol. 2006;31(6):650–60. https://doi.org/10.1093/jpepsy/jsj088 .
Al Shaikh A, Al Zahrani AM, Qari YH, et al. Quality of life in children with diabetes treated with insulin pump compared with multiple daily injections in tertiary care center. Clin Med Insights Endocrinol Diabetes;2020;13:117955142095907. https://doi.org/10.1177/1179551420959077
Müller-Godeffroy E, Treichel S, Wagner VM, on behalf of the German Working Group for Paediatric Pump Therapy. Investigation of quality of life and family burden issues during insulin pump therapy in children with Type 1 diabetes mellitus-a large-scale multicentre pilot study. Diabetic Medicine. 2009;26(5):493–501. doi: https://doi.org/10.1111/j.1464-5491.2009.02707.x
Alqambar M. Impact of insulin pump use on disease control and quality of life in type I diabetes patients. Endocrinol Metab Int J. 2018. https://doi.org/10.15406/emij.2018.06.00169 .
Mesbah NI, Taha NAER, Rahme ZN, Sukkar FF, Omar DM. Experiences of adults using continuous subcutaneous insulin infusion: a qualitative study. Med Princ Pract. 2020;29(3):255–61. https://doi.org/10.1159/000503705 .
Alvarado-Martel D, Velasco R, Sánchez-Hernández RM, Carrillo A, Nóvoa FJ, Wägner AM. Quality of life and type 1 diabetes: a study assessing patients’ perceptions and self-management needs. Patient Preference and Adherence; 2015:1315. https://doi.org/10.2147/PPA.S87310
Undén A-L, Elofsson S, Andréasson A, Hillered E, Eriksson I, Brismar K. Gender differences in self-rated health, quality of life, quality of care, and metabolic control in patients with diabetes. Gender Med. 2008;5(2):162–80.
The authors acknowledge the help of the LatDiane cohort database, the Children’s Clinical University Hospital, the Latvian Association of Endocrinology, and Latvian diabetes organizations.
This study was partially financially supported by Medtronic B.V. Representative Office in Latvia.
Authors and affiliations.
Faculty of Medicine, University of Latvia, Jelgavas Str. 3, Riga, Latvia
Lilian Tzivian & Jelizaveta Sokolovska
Faculty of Humanities, University of Latvia, Riga, Latvia
Anna E. Grike
Faculty of Medicine, Riga Stardins University, Riga, Latvia
Longenesis Ltd, Riga, Latvia
Agate Kalcenaua, Martins Mednis, Ieva Danovska, Ugis Berzins, Arnolds Bogdanovs & Emil Syundyukov
Questrom Business School, Boston University, Boston, MA, 02215, USA
Digital Business Institute, Health Analytics and Digital Health, Boston University, Boston, MA, 02215, USA
Faculty of Industrial Engineering and Technology Management, Holon Institute of Technology, 5810201, Holon, Israel
Faculty of Digital Technologies in Medicine, Holon Institute of Technology, 5810201, Holon, Israel
Faculty of Computing, University of Latvia, Raina boulevard 19, Riga, 1050, Latvia
You can also search for this author in PubMed Google Scholar
LT, JS, ES and AK designed the study; AEG performed face-to-face interviews and analyzed a qualitative part of the study; LT analyzed a quantitative part of the study and wrote the main manuscript text; AS, AB and ES consulted the overall design of the study and wrote the main manuscript text; ID and ES prepared all figures and contributed to development of the digital platform; MM, UB, AB and ES developed a digital platform for the study. All authors reviewed the manuscript. All authors read and approved the final manuscript.
Correspondence to Lilian Tzivian .
Ethics approval and consent to participate.
The study was approved by the Scientific Research Ethic Commission of the Institute of Cardiology and Regenerative Medicine of the University of Latvia on February 2, 2021. Participants provide their consent (which could be dynamically managed on the platform, e.g., for opt-out) whilst entering the Digital Engagement platform.
Consent for publication
Authors declare no conflict of interests.
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Additional file 1.
. Surveys by blocks, qualitative responses of participants, supplemental tables and figures.
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/ . The Creative Commons Public Domain Dedication waiver ( http://creativecommons.org/publicdomain/zero/1.0/ ) applies to the data made available in this article, unless otherwise stated in a credit line to the data.
Reprints and Permissions
About this article
Cite this article.
Tzivian, L., Sokolovska, J., Grike, A.E. et al. Quantitative and qualitative analysis of the quality of life of Type 1 diabetes patients using insulin pumps and of those receiving multiple daily insulin injections. Health Qual Life Outcomes 20 , 120 (2022). https://doi.org/10.1186/s12955-022-02029-2
Received : 05 January 2022
Accepted : 18 July 2022
Published : 01 August 2022
DOI : https://doi.org/10.1186/s12955-022-02029-2
Share this article
Anyone you share the following link with will be able to read this content:
Sorry, a shareable link is not currently available for this article.
Provided by the Springer Nature SharedIt content-sharing initiative
- Type 1 diabetes mellitus
- Insulin pump
- Multiple daily insulin injections
- Real World Data digital tool
- Diabetes-related expenses
- Comparative effectiveness research
- Health economics
Health and Quality of Life Outcomes
- Submission enquiries: [email protected]