Course of body weight before and after the initiation of insulin therapy in type 2 diabetes mellitus: Retrospective inception cohort study (ZODIAC 58)

Abstract Aims The aim of this study was to explore the effect of insulin treatment initiation on weight by taking weight change prior to initiation into account. Materials and methods We performed an observational retrospective inception cohort study, concerning Dutch primary care. We identified all patients that initiated insulin treatment (n = 7967) and individually matched patients with a reference patient (n = 5213 pairs). We obtained estimated mean weight changes in the five years prior to five years post insulin therapy. We applied linear regression analysis on weight change in the first year after insulin therapy (T0 to T+1), with matched group as primary determinant adjusted for pre‐insulin weight change and additional covariates. Results Estimated mean weight increased in the five consecutive years prior to insulin therapy (−0.23 kg in year T‐5 to T‐4, 0.01 kg in year T‐4 to T‐3, 0.07 kg in year T‐3 to T‐2, 0.24 kg in year T‐2 to T‐1, and 0.46 kg in year T‐1 to T0) and continued to increase in the first year after, that is T0 to T+1, at a slightly lower rate (0.31 ± 3.9 kg). Pre‐insulin weight change had the highest explained variance and was inversely and independently associated with weight change (p < .001). Starting insulin was associated with weight increase, independent of pre‐insulin weight change (β‐adjusted 1.228, p < .001). Stratification revealed that despite having a more or less similar baseline BMI, patients with substantial weight increase showed higher estimated mean BMI’s followed by weight loss pre‐insulin. In matched references, estimated mean weight changes were negative in all years concerning the study period, indicating consistent weight loss. Conclusions Initiation of insulin therapy was independently associated with weight increase; however, overall effect on weight was small and subject to substantial variation. Pre‐insulin weight change is identified as a relatively strong inverse determinant of weight change after insulin initiation.


| INTRODUC TI ON
When dietary measures and oral glucose lowering drugs (OGLDs) fail to sufficiently correct glucose dysregulation, most treatment guidelines for type 2 diabetes (T2DM) advise to start or add either an injectable glucagon-like peptide-1 analogue or insulin therapy. 1 Insulin is widely known and used for its glucose lowering effects, but also is an anabolic hormone affecting lipid and protein metabolism. 2,3 Several reports claim that insulin therapy is associated with weight gain. 4 Moreover, the mechanisms behind weight changes after initiation of insulin therapy are multifaceted and not completely understood. 2 Observational studies on the initiation of insulin therapy in patients with T2DM have shown considerable variations in weight gain between patients, [5][6][7][8][9][10][11][12][13][14] which raises the question whether and to what part weight gain is attributable to insulin, and whether and to what part weight gain is attributable to other characteristics. Some differences in weight gain between insulin regimens and types were reported, 6,9,12,13 but also large standard deviations dwarfing these differences. 9 Studies have unanimously reported an inverse association of baseline BMI with weight change after initiation. 6,8,[10][11][12][13][14] Results on baseline HbA1c vary from not significantly 7,13 to significantly 6,8,12 associated with weight change. Other reported variables include HbA1c change, limited to the first nine months 10 or the first year 6 after initiation, and insulin dose and HbA1c at follow-up. 6 No strong baseline determinants have been found. 10,13 The performance of prediction models increased when variables gathered after the decision to start insulin therapy were included. 6,15 Few studies included weight change prior to the initiation of insulin therapy. 9,16,17 To the best of our knowledge, there are currently no large observational intervention studies, specifically studying long-term information on body weight prior to the initiation of insulin therapy in primary care.
We therefore explored the effect of initiation of insulin therapy on body weight by taking weight change prior to initiation into account, using the available data of the Zwolle Outpatient Diabetes project Integrating Available Care (ZODIAC) cohort.

| Setting and participants
This study is part of the studies in the ZODIAC project, performed in a prospective primary care T2DM patient cohort in the Netherlands, initiated in 1998 for benchmarking purposes. 18 Exclusion criteria were insufficient cognitive capabilities or a very short life expectancy, based on the judgement of the general practitioners.
More than 99.5% of patients also consented with the use of their anonymized data for research purposes. Patients included in the ZODIAC cohort were diagnosed with T2DM and treated in primary care, according to national guidelines, that is the Dutch College of General Practitioner Guideline (Dutch: NHG-Standaard). 19 The ZODIAC project was approved by the Ethics Committee of Isala, Zwolle (references 03.0316 and 07.0335).

| Study design
Using the available data in the ZODIAC database of the years 1998 to 2014, we performed a retrospective observational intervention study on the effects of the initiation of insulin treatment on body weight, by means of a new user design/ inception cohort. 20 Figure S1 schematically shows the selection of patients and data for this study. Of all ZODIAC participants, we identified patients that started insulin therapy. Besides extracting data at T0, which is the first registered insulin use signal defined as index time point, we also extracted data of the 5 years prior to T0 and the 5 years after T0.
In addition, a matched reference group of ZODIAC participants who did not start insulin therapy during the study period was selected. Insulin users were individually matched with a referent T2DM patient based on sex, age (±1 year), diabetes duration (±1 year) and BMI-field (±0.5 kg/m 2 ) at T0 in a 1:1 ratio.

| Data sources and measurement
A data set of quality indicators on T2DM care was collected annually by general practices. This core data set included patient demographics, laboratory results, medication use, lifestyle and variables collected through physical examination.
HbA1c and lipid profile were determined using standard laboratory procedures. For this study, HbA1c measurement units were aligned to allow proper comparison. HbA1c measured in DCCT/ NGSP units (%) was converted to IFCC units (mmol/mol) using the following formula: HbA1c mmol/mol = (10.93 * HbA1c %) - 23.5. 21 HbA1c was measured to the nearest mmol/mol. Weight was measured to the nearest kilogram (kg). In the case of a missing BMI-field, BMI (kg/m 2 ) was calculated from weight and height recordings.
Lifestyle factors were self-reported. Physical activity was considered adequate when reported as being performed at least five times a week with a duration of thirty minutes. Clinically unlikely data points were excluded from analysis, as described in Table S1 ) of weight change concerning the time period T-5 to T+5 was visualized, in order to explore changes before and after insulin therapy initiation. Weight change in the first year after index was the primary outcome concerning regression analysis.

| Exposure
The exposure of interest was initiation of insulin treatment, which was represented by the insulin subgroup. Non-exposure was represented by the matched references.

| Covariates
Pre-insulin weight change was studied as main covariate. For this purpose, several measures of weight change concerning the time periods T-3 to T0, T-2 to T0 and T-1 to T0 were studied. Both weight changes in the pre-insulin time periods (eg T0 minus T-2) and averaged weight changes of consecutive years within the time periods (eg T-1 minus T-2, and T0 minus T-1, divided by 2) were calculated.

| Stratified analysis
Longitudinal courses were stratified by weight change category in the first year after index. Weight change was categorized into three categories: 1] substantial weight gain (≥5 kg), 6 2] moderate weight gain (1 to < 5 kg) and 3] weight gain less than 1 kg, that is either mild gain, no change or weight loss.

| Statistical methods
Analysis sets were as follows: The full analysis set (FAS), containing all patients identified to have started insulin therapy in the study period. The intention-to-treat (ITT) analysis set, containing all patients with available weight change data in the first year after index, whether or not they continued insulin therapy. 6,13 The per-protocol (PP) analysis set, containing patients with available weight change data that continued insulin therapy at T+1. Patients were included in the PP analysis set whether or not they changed to a different insulin regime. Finally, patients with available weight change data without insulin use at T+1 were indicated as those that used insulin for less than one year (U < 1y).
Categorical data were presented by n (%), and quantitative variables were presented by mean with standard deviation (SD) or median (Q1 -Q3) depending on the distribution.
Estimated means with 95%CIs of weight change, weight, BMI and HbA1c were obtained through linear mixed model analyses for repeated measures which allows for extrapolation of missing values. 16 Weight change, weight, HbA1c and BMI were dependent variables, and time was a fixed factor. Akaike's information criterion was used to select covariance structures.
We performed linear regression analysis to assess the effect of initiation of insulin treatment on weight change, relative to matched references. The dependent variable was weight change (kg) in the first year after index, and matched study group, pre-insulin weight change and additional covariates were studied as independent variables.
We performed sensitivity analysis concerning weight change in the year prior to index. This was added due to the fact that insulin therapy was initiated in the time period T-1 to T0, but inherent to a new users design the exact timing of insulin initiation could not be identified. 20 Analyses were performed using RStudio version 1.

| Participants
We identified n = 7967 patients that started insulin therapy during the study period. None of the patients used insulin before T0.
For n = 5213 patients that started insulin therapy an individually matched reference patient was available. Table 1  The insulin subgroup and matched reference group were comparable regarding sex, age, diabetes duration and BMI. These groups were different regarding HbA1c, as HbA1c is a main driver of the decision to start insulin therapy. Mean HbA1c was 58.4 (±11.2) mmol/mol for the insulin subgroup and 48.9 (±8.5) mmol/mol for the matched references, resulting in a mean difference of 9.5 mmol/mol.

| Analysis sets
Weight change data in the first year after index were available for n = 5086 patients (ITT analysis set), of which n = 4291 patients also   (Table S2).
Availability of weight change data differed for the matched groups. These were available for n = 3433 and n = 3511 patients of the insulin subgroup (ITT-insulin set) and matched references (ITTreference set), respectively. Weight change data were available for n = 2812 matched pairs, referred to as the ITT-matched analysis set.
Of these, the insulin subgroup of n = 2349 matched pairs still used insulin at T+1, referred to as the PP-matched analysis set. In contrast to the insulin subgroup, estimated mean weight changes in the matched reference group were negative in all years concerning the study period, which indicates consistent mean weight loss.     Note: One-way ANOVA and post-hoc analyses, using Bonferroni-adjusted p-values, were performed: In the complete insulin group, there was no significant difference between insulin category, whereas there was between BMI category (p < .001 for ITT and PP) and HbA1c tertile (p = .005 for ITT and p = .003 for PP).

| Weight change in the first year
In the insulin subgroup, there was no significant difference between insulin category, whereas there was between BMI category (p

| Effect of insulin therapy on body weight
Based on univariate analysis, weight change T-2 to T0 in kg was chosen as main covariate (Table S4, Figure S3). In both the insulin subgroup and matched references, weight change T-2 to T0 was inversely associated with weight change in the first year after index.

| Stratified analysis
In the first year after index, substantial weight gain was seen in n = 509 patients (10.0%), moderate weight gain was seen in n = 1916 patients (37.7%), and mild gain/ no change/ weight loss was seen in n = 2661 patients (52.3%). Figure 2 shows that patients with substan-

| D ISCUSS I ON
This study showed that estimated mean weight gradually increased in the years prior to insulin therapy and continued to increase in the first year after, albeit at a slightly lower rate. Pre-insulin weight change was inversely associated with weight change in the first year after, with the highest statistical explained variance of all modelled covariates. Initiation of insulin therapy was associated with weight increase, independent of pre-insulin weight change and additional covariates. Moreover, compared to the ITT-matched analysis (assessment of treatment policy), the associations in the PP-matched analysis (assessment of the true effect of a drug) were the strongest.
Mean weight gain in the first year after insulin therapy, however, was small and subject to substantial variation. Patients with substantial weight gain showed high initial pre-insulin estimated mean weight/ BMI values, and mean weight/ BMI loss in the year preceding insulin.
Few other studies included pre-insulin weight change. 9,16,17 Of these, Gordon et al 9 visualized a more or less stable weight prior to insulin therapy. 9 Another observational study by Gant et al, 16 in not necessarily naïve insulin users referred to secondary care, reported a similar-shaped curve concerning averaged BMI course. 16 The averaged BMI course did not significantly differ between insulin users and non-insulin users (p-interaction = 0.460).
In the present study, we adjusted insulin therapy initiation for pre-insulin weight change, showing that insulin therapy is independently associated with weight increase in the first year after. In the study by Larger et al, 17 all patients lost weight prior to insulin therapy due to unreported cause, followed by initiation of insulin therapy with delineating weight gain. 17 Weight gain after insulin therapy initiation was correlated with prior maximum lifetime weight and maximal daily insulin dose, 17 resulting in the hypothesis that weight gain observed after the introduction of insulin may mostly be catch-up weight gain. 26 In the present study, by stratification, we showed that only those patients with substantial weight gain had estimated mean weight loss in the year preceding insulin. This subgroup showed an estimated mean BMI decrease, which at T0 reached a value within the range of the other subgroups. Combined with an unaltered high estimated mean HbA1c at T-1 and T0, these results may indicate poor glucose control with subsequent increased glycosuria resulting in weight loss. 2 An alternative explanation for pre-insulin weight loss could be that, by losing weight, patients attempted to postpone insulin therapy. However, pre-insulin estimated mean weight loss was not found in the other categories. Concerning matched references, attempting to postpone insulin therapy could explain the finding that patients in the highest HbA1c tertile lost the most weight.
Determinants of substantial weight gain after the initiation of insulin therapy will be further investigated in the second stage of this study.
Several studies, including the present one, found an inverse association of baseline weight/ BMI with weight change after initiation. 6,8,[10][11][12][13][14] In contrast with general clinical belief, reassurance with regard to the use of insulin in obese patients was reported. 6,11 In this study, we showed that patients with substantial weight gain had a more or less similar estimated mean BMI at baseline, but markedly higher estimated BMI values prior to baseline (T-5 to T-1). This novel finding indicates that using baseline BMI as a determinant of weight gain after insulin therapy initiation may be misleading.

| Mean weight gain after insulin initiation
A mean weight gain of 0.31 kg to 0.43 kg in the first year after initiation of insulin therapy is small as compared to several others reports. Of presented studies, 5-14 mean one-year weight gains of 0.98 ± 7.1 kg, 7 1.2 kg, 10 1.78 to 2.00 kg, 5,6 2.8 ± 6.0 kg 9 and 3 kg 13 were reported.
Differences in weight gain may be attributable to various, whether or not coherent, factors, for example patient characteristics, study medication including dose, co-medication, analysis sets and study setting. The CREDIT study, a multinational study, showed that mean weight gain substantially differs between countries, ranging from 0.95 kg in Germany to 4.26 kg in Portugal. 6 Also, regional differences within the United Kingdom were reported. 12

| The Netherlands
In the Netherlands, everyone has a general practitioner and over 80% of patients with T2DM is treated in primary care. Patients were treated according to the Dutch College of General Practitioner Guideline of the study period, advising to start insulin therapy when HbA1c is at 53 mmol/mol (7%), allowing adjustment with higher cutoff points in the elderly and frail. 19

| Generalizability
In general, the results found in ZODIAC are considered representative for the Netherlands, and therefore generalizable, at least for

| Strengths and limitations
Strengths and limitations inherent to the study design were previously described. 6,20,23 The effect of insulin was studied in routine clinical practice, which allows for real-life information on use in the general population. As data were gathered annually, the exact time of insulin therapy initiation could not be identified. Also, we cannot exclude the possibility of some registration delay, which we believe has a maximum of one year. Furthermore, we cannot exclude the possibility that some patients in the insulin-only group were also using OGLDs; chances are that this will only be the case in a minority of patients. The clinical characteristics of the insulin-only users were barely different from the patients treated with a combination of insulin and OGLDs (data not shown).
Insulin doses, frequency of injections and nutritional factors were not included in ZODIAC. Also, study site was not taken into account.
Weight was measured to the nearest kg, hence more as a discrete than a continuous variable. Body fat distribution was not measured. 27 Furthermore, weight change cannot be attributed to fat tissue only, as an unmeasured fat free component notably fluid may play a role in weight change. 24 Moreover, no patient selections other than those described in the methods section were made; hence, the group also included some patients with clinically relevant renal disease (a rare patient group in the primary care in the Netherlands) and patients with chronic heart failure. Since no structural information on fluid status or degree of chronic heart failure was available, the influence of these factors could not be taken into account.

| Conclusion
We conclude that initiation of insulin therapy was independently associated with weight increase; however, overall effect on weight was small and subject to substantial variation. Pre-insulin weight change is identified as a relatively strong inverse determinant of weight change after insulin initiation.

CO N FLI C T O F I NTE R E S T
The authors have nothing to declare.

AUTH O R CO NTR I B UTI O N S
MAE obtained the idea for the study, designed the study, researched data and wrote the manuscript, PRvD contributed to the discussion and reviewed/edited the manuscript, EH co-designed the study, contributed to the discussion and reviewed/edited the manuscript, and HJGB obtained the idea for the study, designed the study, contributed to the discussion and reviewed/edited the manuscript.

DATA AVA I L A B I L I T Y S TAT E M E N T
Data available on request from the authors.