Integrated care for type 1 diabetes: The West Bengal model: 24‐month follow‐up

Abstract

Aims

This study aimed to document outcomes of care for individuals with type 1 diabetes (T1D) facilitated by a government‐funded model supporting establishment of clinics within secondary‐level district hospitals in a lower‐middle‐income country.

Methods

This prospective, multicentre, single‐arm pre‐post implementation evaluation study was conducted in government‐funded ‘model’ clinics in district hospitals in West Bengal, India. All consecutive persons with physician‐diagnosed T1D (N = 366) were enrolled and immediately transitioned to basal–bolus insulin regimen, with structured monthly follow‐up visits. Clinical measures, psychological well‐being and diabetes‐related expenditures were assessed at baseline, 12 months and 24 months.

Results

At baseline, median age and age at diagnosis were 15 (10–21) and 8 (5–11) years, respectively, with 41.4% from lower socio‐economic class, and most parents/caregivers had not completed primary education. There were no dropouts, with a mean of 23.5 follow‐up visits over 24 months. Median glycated haemoglobin (HbA1c) fell from 79 mmol/mol (64–104) [9.4% (8.0–11.7)] at baseline to 72 mmol/mol (58–87) [8.7% (7.5–10.1)] at 12 months, which further fell to 64 mmol/mol (54–73) [8.0% (7.1–8.8)] at 24 months (p < 0.0001). No episodes of documented diabetic ketoacidosis, hospital admissions or deaths occurred. Psychological well‐being improved. Median monthly expenditures dropped from 2600 INR (~30 USD) at baseline to 200 INR (~2 USD) at 24 months (p < 0.0001).

Conclusions

This novel public health system‐supported model of T1D improved clinical parameters and psychological well‐being of those living with T1D and their parents/caregivers and reduced expenditures.

What’s new?

What is already known?

• There is a lack of data related to outcomes of government‐funded structured care model for type 1 diabetes at secondary care level in low‐ and middle‐income countries.

What this study has found?

• The model led to improvements in clinical parameters, improved psychological well‐being and reduced family type 1 diabetes‐related expenses, despite most persons being of lower socio‐economic status.

What are the implications of the study?

• The model can guide development of national health policy to improve care for type 1 diabetes by demonstrating a scalable, cost‐effective, decentralized approach. The model is adaptable to other low‐ and middle‐income countries.

1. INTRODUCTION

Management of Type 1 diabetes (T1D) requires a complex multidisciplinary multimodality approach including multiple daily insulin injections, self‐monitoring of blood glucose (SMBG), prevention and management of acute and chronic diabetes‐related complications, healthy meal plan and regular physical activity. ¹ Very few public health systems in low‐ and middle‐income countries are able to provide this level of care at secondary‐level health facilities (district hospitals), leaving a critical gap in structured T1D management. ² , ³ , ⁴ T1D is not included in any health programs/policies in India ⁵ although, partly due to its large population, India has the highest number of children and adolescents with T1D globally (estimated at 307,000 in 2025). ⁶ Challenges in timely diagnosis and lack of structured care lead to premature death: it is estimated that an Indian child diagnosed with T1D aged 10 years in 2025 has an average of only 30 years additional life expectancy. ⁶ A single‐centre study from an Indian tertiary care hospital reported a high mortality rate of 11.6% over mean follow‐up of 6.4 years among 898 individuals with youth‐onset T1D, of whom 75.6% had HbA1c > 86 mmol/mol (10%). ⁷

In West Bengal, we conducted a gap‐analysis study of clinical outcomes in children and adolescents with T1D receiving unstructured care in a tertiary care hospital and found poor glycaemic control, poor adherence to insulin regimen, irregular home blood glucose monitoring and high frequency of complications. ⁸ Under the aegis of Government of West Bengal, India, we took the initiative to develop, implement and evaluate a structured model of care for T1D in secondary care‐level district hospitals of five health districts of West Bengal in order to improve physical, social and psychological well‐being and reduce financial burden for children and adolescents living with T1D. The model is based on ‘Intermediate Care’, the suggested standard of care for T1D in low‐ and middle‐income countries ³ , ⁹ with key features basal–bolus insulin regimen, diabetes education, two to four SMBG tests per day, glycated haemoglobin (HbA1c) testing every three months and annual screening for diabetes‐related complications. ³ , ⁹ In India, the district hospital is at secondary referral level. Bed strength varies from 75 to 500 beds depending on the size, terrain and population of the district. ¹⁰ The Life for a Child program of Diabetes Australia has been providing technical advice. ¹¹ Some details of the method have been published previously. ¹²

This study aimed to evaluate the impact of the implementation of the model on clinical parameters of the participants, psychological well‐being of participants and their parents/caregivers, and direct and indirect T1D‐related expenditure incurred by families at 12 months and at 24 months.

2. METHODS

2.1. Study background

Dedicated T1D clinics were established in the first five district hospitals, and subjects from these clinics were subsequently enrolled in the study. Clinics were established by leveraging the existing health care delivery system. This was complemented by intensive training and mentoring of dedicated health professionals and the provision of essential supplies. These supplies included insulin (basal–bolus regimen, including insulin glargine and human regular insulin), blood glucose meters, test strips (100 per month), lancing devices, lancets (50 per month), insulin syringes (31‐gauge needles) (40 IU 30 per month and 100 IU 10 per month) and laboratory investigations. We established an electronic person registry with real‐time electronic and physical prescription and data capture to facilitate continuous auditing (e.g., regular data quality checks and process reviews) of processes and outcomes and ongoing clinical follow‐up. ¹² , ¹³ The clinics included a physical space with desks, chairs, weighing machine, stadiometer, sphygmomanometer and computer with internet services as part of the NCD adult clinic infrastructure. Existing adult Non‐Communicable Disease (NCD) clinics of district hospitals have been upgraded to run as dedicated T1D clinics once a week, utilizing existing infrastructure and manpower. ¹² , ¹³

Specific T1D clinic related resources for implementing the T1D clinic model included procurement of refrigerators for storing insulin and drugs, and cupboards for storage of blood glucose meters and strips, lancing devices, lancets, insulin syringes, education materials, cards, forms, etc. Education materials included: structured age‐appropriate diabetes self‐management education booklets developed in local languages including insulin administration, injection site rotation, hypoglycaemia management, sick‐day rules, home blood glucose monitoring and meal planning; illustrated posters and flip charts; logbooks for blood glucose monitoring and insulin dose titration; and emergency action cards for hypoglycaemia and diabetic ketoacidosis. The model was designed to enable dispensing of consumables and supplies from the clinic itself rather than the main store and pharmacy of the hospital. This was done to enable maintenance of a steady supply chain and ensure accountability and auditing of supplies.

Dedicated clinicians (physician and/or paediatrician), clinic nurse, NCD counsellor and data entry operator at district hospital level were rigorously trained to deliver quality care for T1D and to facilitate ongoing monitoring and evaluation of the model. The majority of healthcare professionals did not have prior experience in managing T1D clinics. Hence, mentoring was done for a period of six months by qualified endocrinologists, who provided inputs and advice and suggested course correction measures without actually being involved in person consultation. The mentoring process included weekly on‐site visits during the initial six months. Guidance focused on insulin titration, glycaemic pattern review, hypoglycaemia prevention, sick‐day management and complication screening. The model is described in further detail in Ghosh et al. ¹² The model was designed to provide ‘intermediate level of care’ which if implemented appropriately, it is anticipated it would lead to HbA1c lowering with target mean HbA1c levels as low as 64 mmol/mol (8.0%). ³ At enrolment, all study participants who were not already on basal–bolus regimen, were transitioned to basal–bolus insulin regimen as part of the intervention model.

This study was designed as implementation research evaluation to assess clinical outcomes of decentralized T1D care in district hospitals in India, rather than a comparative effectiveness trial. By integrating an implementation science framework such as the Consolidated Framework for Implementation Research (CFIR) and principles outlined by Yapa et al., ¹⁴ we aimed to operationalize dedicated T1D clinics with identified human resources, rigorous training, mentoring and steady supply chain for insulin and essential commodities. Implementation strategies included training general physicians, nurses and NCD counsellors; monthly person follow‐up through physical visits; and annual screening for complications. Mandatory monthly follow‐up of all subjects in physical mode was an integral component of the intervention model, designed to enable close monitoring of glycaemic patterns, appropriate insulin titration, providing diabetes self‐management education and screening for complications. This was implemented as part of routine service delivery rather than as a requirement for research. All essential supplies including insulin and other consumables were provided to the subjects on a monthly basis.

2.2. Funding information

The intervention was supported and funded by the Government of West Bengal and part‐funded by the Government of India, via a Program Implementation Plan (NHM Administrative Approval FY 2022–23 & FY 2023–24_West Bengal, Page 173 under scheme/activity: State specific programme interventions and innovations), through the National Health Mission. The State Non‐Communicable Disease (NCD) Directorate was responsible for regular visits and continuous audit of processes and outcomes. Importantly, the model was designed to piggyback on existing infrastructure and personnel. UNICEF is a project partner since December 2024.

2.3. Ethical considerations

For evaluation of this model, ethical clearance was obtained from the Institutional Ethics Committee of the Institute of Post‐Graduate Medical Education and Research, Kolkata, India (Memo No. IPGME&R/IEC/053R). The research was conducted in compliance with the Declaration of Helsinki. Written informed consent was obtained from adult participants and assent from child or adolescent participants as well as written informed consent from their parents/caregivers. The Institutional Ethics Committee determined that a randomized controlled trial (RCT) would be unethical (Memo No. IPGME&R/IEC/0862) considering poor clinical outcomes and high mortality rate among those with T1D in India. ⁷ , ⁸ Additionally, since services were provided by the government, it would be unethical to deprive possible benefits to any subject. Therefore, a control group was not possible.

2.4. Sample size calculation

HbA1c was chosen as primary end point for power calculation. Conventionally, HbA1c lowering effect of any intervention is considered clinically significant when there is a change of 4 mmol/mol (0.4%) from baseline. ¹⁵ , ¹⁶ Assuming an average HbA1c of T1D persons in the Indian population as 69 ± 15 mmol/mol (8.5% ± 1.4%), ¹⁷ higher power of 99% and stringent α error (two‐sided) of 1% to reduce Type I and Type II errors, it was calculated that at least 332 subjects are necessary in order to achieve a reduction of 4 mmol/mol (0.4%) in HbA1c. Adding a non‐response rate of 10%, the final sample size was calculated to be 366.

2.5. Participants

Two groups of subjects were enrolled as beneficiaries under the model. The first group included subjects with T1D who were already being treated elsewhere (predominantly in tertiary care public health facilities) and were shifted to their respective five home districts, and secondly, subjects who were diagnosed at the district hospital level after the establishment of the model clinics. We excluded the second category of subjects from analysis because recent onset/diagnosis of disease would affect HbA1c results (Figure 1). The final sample of 366 study participants (approximately 73 from each of the five district hospitals) included only those with established T1D who were already being treated elsewhere and were subsequently shifted to the model clinics, enrolled by consecutive sampling technique.

FIGURE 1.

Study flow diagram.

Baseline parameters were compared with that at 12 months and at 24 months.

2.6. Diagnosis of type 1 diabetes

Diagnosis was made on the basis of diagnostic criteria for diabetes (American Diabetes Association 2022). ¹⁸ Autoantibody testing was not done routinely for all study participants as part of the model due to resource constraints within the public health setting. However, in a previously published study, which included part of the current cohort (92 T1D subjects from one hospital), it was found that 97.8% had at least one T1D autoantibody positive and 67.3% had at least two T1D autoantibodies positive, ¹⁹ supporting the accuracy of clinical diagnosis.

2.7. Procedures

2.7.1. Socio‐demographic characteristics and relevant history of study participants

We recorded socio‐demographic details of study participants including age at baseline, sex, current status (in school/college/vocational training/employment/others), educational qualification of parents, occupation of parents, monthly family income (Indian Rupees (INR); 1 US$ = INR 87.62 as on 8 August 2025), ²⁰ per capita monthly family income (INR) and social class in a proforma. Educational qualification of parents was divided into six categories: no schooling or completing < 5 years, 5–7 years, 8–9 years, 10–11 years and 12 or more years. ²¹ Occupation of parents was divided into seven categories: professional (technical/administrative/managerial), clerical, sales worker, agricultural worker, service worker, production worker (skilled/unskilled manual worker) and not employed. ²¹ Social class of study participants was divided into five categories according to Modified B. G. Prasad Classification 2023: upper class (9098 and above), upper middle class (4549–9097), middle class (2729–4550), lower middle class (1365–2728) and lower class (below 1365). ²² , ²³

Relevant history of study participants was also obtained including age at diagnosis of diabetes, duration of diabetes, presentation at diagnosis, family history of diabetes (first‐degree relative diagnosed with any form of diabetes) and presence of acanthosis nigricans or goitre.

Data related to regimen of insulin treatment, insulin injection sites, frequency of SMBG in last week, and whether SMBG was done by self or parent/caregiver were obtained at baseline and at 24 months. We documented whether subjects were on basal–bolus, split‐mix or premixed insulin regimen at baseline and follow‐up. We also performed a sub‐group analysis of outcomes based on whether study participants were on basal–bolus insulin at baseline or not. This allowed assessment of whether prior regimen influenced HbA1c improvement. Diabetes‐related acute complications (diabetic ketoacidosis) rates were documented in the one year preceding enrolment and for 12 months and 24 months following enrolment while hypoglycaemia was documented in the two months preceding enrolment and for 12 months and 24 months following enrolment.

Diabetic ketoacidosis was defined as a triad of hyperglycaemia, ketosis and metabolic acidosis as per international guidelines. ²⁴ Overall symptomatic and/or documented hypoglycaemia capillary plasma glucose <70 mg/dL (3.9 mmol/L) included all episodes of plasma glucose concentration low enough to cause symptoms and/or signs and expose the individual to potential harm. ²⁵ Injection sites were examined and categorized as healthy/non‐healthy.

2.7.2. Evaluation of clinical parameters of study participants

Blood samples were analysed by standard methods at baseline, after 12 months and after 24 months for fasting plasma glucose (FPG); post‐prandial plasma glucose (PPPG); HbA1c (minimum four readings in 24 months, apart from baseline results); fasting lipid profile including total cholesterol, triglycerides, high‐density lipoprotein (HDL) cholesterol and low‐density lipoprotein (LDL) cholesterol; thyroid stimulating hormone (TSH) and free thyroxine (FT4). HbA1c estimation was done at the biochemistry laboratory of each district hospital by High Performance Liquid Chromatography (HPLC) using Bio‐Rad D‐10 (Bio‐Rad Laboratories, Inc., Hercules, CA, USA), and other biochemical variables were measured by standard techniques.

Height and weight were measured by standard methods. Participants were divided into three age groups: < 10 years, 10–15 years and > 15 years.

2.7.3. Evaluation of psychological well‐being of study participants and their caregivers

Psychological well‐being of study participants and their caregivers was assessed at baseline and at 12 months and 24 months using validated questionnaires with validated translations in local languages, which were used previously in studies in India. Diabetes‐specific health‐related quality of life (HRQOL) was assessed using Peds QL^TM 3.2 Diabetes Module, ²⁶ , ²⁷ , ²⁸ general well‐being was assessed using Peds QL^TM 4.0 Generic Core Scales, ²⁹ , ³⁰ , ³¹ and family functioning and social support were assessed using Peds QL^TM Family Impact Module Version 2.0. ³² The different language versions of the Peds QL™ used in this study were obtained from Mapi Research Institute, which follows standardized protocol for country‐specific adaptations of QOL questionnaires. Necessary permissions were taken from appropriate authorities for using these questionnaires. Questionnaires were self‐administered for parents/ caregivers and study participants aged ≥ 8 years. In the age group of 5–7 years, parents assisted the children in completing the questionnaires, and parents completed the questionnaire for those aged 2–4 years. Study investigators read the questions out when the parents/caregivers were illiterate.

2.7.4. Evaluation of direct and indirect expenditure incurred by families

Direct and indirect expenditure incurred by families in the month preceding enrolment was documented. Direct expenditure incurred by families included costs of insulin, glucose measuring devices and blood glucose test strips, insulin syringes and laboratory investigations. Indirect expenditure incurred by families included costs of transport to and from hospital, food and loss of wages. Recall period of one month was used to obtain information regarding direct and indirect expenditure incurred by families of study participants.

2.8. Statistical analysis

Study participants were enrolled in an online system based on Aadhaar number (national identification number) which prevented data duplication. Statistical analyses were performed using Microsoft Excel and SPSS for Windows version 21.0; Inc. Chicago, IL, USA. Data were assessed for normality using the Kolmogorov Smirnov test and Q‐Q plots. Continuous variables were categorized as normally distributed variables and non‐normally distributed variables. Normally distributed variables were presented as mean ± SD and non‐normally distributed variables as median with interquartile range (IQR). For non‐normally distributed variables, appropriate non‐parametric tests were conducted. Differences in repeated measures across three time points were analysed using the Friedman test. If significant, it was followed by post‐hoc pairwise comparisons (baseline vs. 12 months, 12 months vs. 24 months and baseline vs. 24 months) using Wilcoxon signed‐rank tests with Holm adjustment applied to control family‐wise error rate due to multiple comparisons. Effect size for pairwise comparison at baseline vs. 24 months was expressed as Hodges‐Lehmann median change with 95% confidence interval (CI). The Hodges‐Lehmann estimator indicates median of all within‐subject pairwise differences between baseline and 24 months, with 95% confidence intervals. It may differ from simple difference in marginal medians. Differences in HbA1c change from baseline to 24 months between insulin regimens were analysed using the Mann–Whitney U test and Hodges‐Lehmann estimator was used to report median change with 95% CIs.

Descriptive statistics were used to summarize baseline characteristics and psychometric properties of the PedsQL^TM instruments. Internal consistency was assessed using Cronbach's α, with values ≥ 0.70 considered acceptable. Floor and ceiling effects were calculated as the percentage of respondents with lowest and highest possible scores, respectively, with < 15% indicating adequate score dispersion. In order to document changes in health‐related quality of life (HRQOL), linear mixed‐effects models were applied with random intercept for each participant and fixed effect for time (baseline, 12 months, 24 months). Models were fitted separately for each total scale score, summary score and subscale score. Estimated marginal means (and 95% CIs) were documented from the models, and mean change from baseline to 24 months was calculated. Statistical significance was tested for three pairwise contrasts (baseline vs. 12 months, baseline vs. 24 months, 12 months vs. 24 months), with Holm adjustment for multiple comparisons. Results were compared with minimal clinically important differences (MCIDs) scores based on published thresholds from Varni et al. (2018) ²⁶ for the PedsQL^TM 3.2 Diabetes Module and Varni et al. ³⁰ for the PedsQL^TM 4.0 Generic Core Scales (≥ 5‐point change for total and summary scores). No published MCIDs exist for PedsQL™ Family Impact Module and hence, changes in these scores were interpreted descriptively, with higher scores indicating better functioning. All analyses were conducted using complete available data at each timepoint. Robustness check (sensitivity analysis) was performed on complete‐case cohort using Friedman/Wilcoxon tests with Holm adjustment for multiple comparisons. Robustness was inferred when both primary linear mixed‐effects model and non‐parametric analysis demonstrated consistent effect direction, magnitude and statistical significance.

Categorical variables were expressed as frequencies and percentages. Binary outcomes at baseline, 12 months and 24 months were summarized as n (%). Proportions across all three timepoints were compared using Cochran's Q test, followed by pairwise comparisons with McNemar's exact test. Absolute risk difference (95% CI) was calculated between baseline and 24 months. p < 0.05 was regarded as statistically significant.

3. RESULTS

3.1. Socio‐demographic characteristics and relevant history of study participants

Number of study participants was 366 (all eligible participants took part in the study consecutively), of whom 192 (52.5%) were women and 174 (47.5%) were men. At enrolment, median participant age was 15 (10–21) years. Median age at diagnosis of diabetes was 8 (5–11) years and median duration was 6 (2–11) years. At diagnosis, 52.5% had presented with diabetic ketoacidosis (DKA), another 42.0% were diagnosed from osmotic symptoms alone, and 5.5% were diagnosed from other symptoms. Socio‐demographic characteristics and relevant history of study participants are shown in Table 1.

TABLE 1.

Socio‐demographic characteristics and relevant history of study participants (N = 366).

Age, years Median (IQR)	15 (10–21)
Sex n (%)
Men	174 (47.5%)
Women	192 (52.5%)
Age at diagnosis of diabetes, years Median (IQR)	8 (5–11)
Duration of diabetes, years Median (IQR)	6 (2–11)
Presentation at diagnosis n (%)
Diabetic ketoacidosis	192 (52.5%)
Osmotic symptoms alone	154 (42.0%)
Others	20 (5.5%)
Family history of diabetes n (%)	60 (16.4%) [12 (3.3%) subjects had family history of T1D]
Acanthosis nigricans n (%)	33 (9.0%)
Goitre n (%)	24 (6.6%)
Current status n (%)
School	219 (59.8%)
College	28 (7.7%)
Vocational training	17 (4.6%)
Employment	38 (10.4%)
Others	64 (17.5%)
Mother's education n (%)/Father's education n (%)
No schooling	66 (17.9%) / 81 (22.1%)
< 5 years complete	35 (9.7%) / 43 (11.7%)
5–7 years complete	45 (12.4%) / 58 (15.9%)
8–9 years complete	63 (17.2%) / 63 (17.2%)
10–11 years complete	104 (28.3%) / 53 (14.5%)
12 or more years complete	53 (14.5%) / 68 (18.6%)
Mother's occupation n (%)/Father's occupation
Professional	0 (0%) / 15 (4.1%)
Clerical	3 (0.7%) / 15 (4.1%)
Sales worker	3 (0.7%) / 88 (24.1%)
Agricultural worker	11 (3.4%) / 48 (13.1%)
Service worker	18 (4.8%) / 20 (5.5%)
Production worker	18 (4.8%) / 141 (38.6%)
Not employed	313 (85.5%) / 39 (10.3%)
Monthly family income, INR a Median (IQR)	7000 (5000–10,000)
Per capita monthly family income, INR a Median (IQR)	1667 (1000–2500)
Social class (modified B. G. Prasad classification 2023) INR a /month n (%)
Upper class (9098 and above)	5 (1.4%)
IIUpper middle class (4549–9097)	15 (4.1%)
IIIMiddle class (2729–4550)	58 (15.9%)
IVLower middle class (1365–2728)	136 (37.2%)
VLower class (below 1365)	152 (41.4%)

^a 1 US$ = INR 87.62 Reserve Bank of India. ²⁰

At baseline, 216 (59.0%) study participants were on basal–bolus regimen of insulin treatment, 129 (35.3%) were on split‐mix regimen, and 21 (5.7%) were on premixed insulin. At enrolment, all study participants who were not already on basal–bolus regimen were transitioned to basal–bolus insulin regimen as part of the intervention model and were rigorously followed up every month through physical visits to the clinic. At the end of two years, the study drop‐out rate was zero, and the mean frequency of follow‐up was 23.5 times in 24 months. Median total daily insulin requirement of study participants decreased from 1.06 units/kg body weight at baseline to 0.94 units/kg body weight at 24 months (p < 0.0001), with a median difference of −0.12 (95% CI: −0.15 to −0.09). Insulin injection sites were healthy in 278 (75.9%) study participants at baseline, which increased to 326 (89.1%) at 12 months and to 362 (98.9%) at 24 months (p < 0.0001). All pairwise comparisons remained significant, with an absolute risk difference of 23.0% (95% CI: 18.8% to 27.7%) between baseline and 24 months. Median frequency of SMBG in the last one week also improved drastically from 2 (0–7) at baseline to 14 (13–15) at 24 months (p < 0.0001). At baseline, 290 (79.2%) study participants performed SMBG themselves, which increased to 312 (85.2%) at 12 months and to 327 (89.3%) at 24 months. At 24 months, only 39 (10.7%) study participants depended on their caregivers for performing SMBG, among whom 31 (8.5%) study participants were below seven years of age.

3.2. Evaluation of clinical parameters of study participants

Median HbA1c of study participants fell from 79 mmol/mol (64–104) [9.4% (8.0–11.7)] at baseline to 72 mmol/mol (58–87) [8.7% (7.5–10.1)] at 12 months, which further fell to 64 mmol/mol (54–73) [8.0% (7.1–8.8)] at 24 months (p < 0.0001 for overall change by Friedman test). Post‐hoc Wilcoxon signed‐rank tests, with Holm adjustment, showed statistically significant reductions at baseline vs. 12 months, 12 months vs. 24 months and baseline vs. 24 months (p < 0.0001). Hodges‐Lehmann median change in HbA1c was −20.77 mmol/mol (95% CI: −25.68 to −15.85), which is equivalent to −1.90% (95% CI: −2.35 to −1.45). At 24 months, 49 (13.4%) study participants had HbA1c ≤ 48 mmol/mol (6.5%), 80 (21.8%) had HbA1c > 48 mmol/mol (6.5%) to ≤ 59 mmol/mol (7.5%), 101 (27.7%) had HbA1c > 59 mmol/mol (7.5%) to ≤ 69 mmol/mol (8.5%), 74 (20.2%) had HbA1c > 69 mmol/mol (8.5%) ≤ 80 mmol/mol (9.5%), and 62 (16.8%) had HbA1c > 80 mmol/mol (9.5%).

Median FPG decreased from 155.0 mg/dL (111.0–204.0) at baseline to 119.0 mg/dL (97.0–158.0) at 12 months and to 105.0 mg/dL (85.8–128.0) at 24 months (p < 0.0001 for overall change), with all pairwise comparisons remaining significant after Holm adjustment. Median change in FPG was −46.0 mg/dL (95% CI: −55.8 to −36.2). Similarly, median PPPG decreased from 192.0 mg/dL (138.0–256.0) at baseline to 131.0 mg/dL (99.0–171.0) at 24 months (p < 0.0001), with an overall median reduction of −51.5 mg/dL (95% CI: −71.4 to −31.6).

Details of clinical parameters of study participants are shown in Table 2.

TABLE 2.

Glycaemic control and other clinical parameters.

Variables	Baseline Median (IQR) (n = 366)	12 Months Median (IQR) (n = 366)	24 Months Median (IQR) (n = 366)	Hodges‐Lehmann Median Change (95% CI) Baseline → 24 Months	Global p a
HbA1c (mmol/mol) HbA1c (%)	79 (64–104) mmol/mol 9.4 (8.0–11.7) %	72 (58–87) mmol/mol* 8.7 (7.5–10.1) %*	64 (54–73) mmol/mol*† 8.0 (7.1–8.8) %*†	−20.77 mmol/mol (95% CI: −25.68 to −15.85) −1.90% (95% CI: −2.35 to −1.45)	< 0.0001
Fasting plasma glucose (FPG), mg/dl	155.0 (111.0–204.0)	119.0 (97.0–158.0)*	105.0 (85.8–128.0)*†	−46.0 (95% CI: −55.8 to −36.2)	< 0.0001
Post‐prandial plasma glucose (PPPG), mg/dl	192.0 (138.0–256.0)	142.0 (109.0–197.0)*	131.0 (99.0–171.0)*†	−51.5 (95% CI: −71.4 to −31.6)	< 0.0001
Total cholesterol, mg/dl	158.0 (135.0–184.0)	164.0 (144.0–183.0)	153.0 (139.0–173.0)	−6.4 (95% CI: −11.7 to −0.9)	0.1
Triglycerides, mg/dl	99.8 (76.8–148.0)	115.0 (80.1–145.0)	101.0 (77.3–142.0)†	+4.4 (95% CI: −3.8 to +12.5)	0.009
High‐density lipoprotein (HDL), mg/dl	51.1 (41.5–59.5)	50.2 (40.0–60.0)	49.7 (44.0–57.9)†	−0.9 (95% CI: −4.3 to +2.6)	0.001
Low‐density lipoprotein (LDL), mg/dl	92.2 (75.2–117.0)	87.3 (71.0–105.0)*	78.5 (66.7–96.8)*†	−14.0 (95% CI: −22.5 to −5.5)	< 0.0001
Thyroid stimulating hormone (TSH), uIU/ml	2.4 (1.6–3.8)	2.5 (1.7–3.7)	3.1 (2.2–4.3)	+0.7 (95% CI: +0.2 to +1.3)	< 0.0001
Free thyroxine (Free T4), ng/dl	1.2 (1.2–1.4)	1.2 (1.0–1.3)	1.2 (1.1–1.3)	−0.05 (95% CI: −0.09 to −0.004)	0.073
Urinary albumin: creatinine ratio, mcg/mg	19.1 (11.0–42.9)	14.2 (8.0–29.6)	11.0 (6.4–21.9)*†	−5.1 (95% CI: −10.6 to +0.5)	< 0.0001

Note: Expressed as Median (Interquartile Range). HbA1c measurements are reported in IFCC units (mmol/mol–no decimal point) followed by DCCT units (%) in parentheses for each time point. Holm‐adjusted Wilcoxon signed‐rank contrasts (one‐tailed): *p < 0.05 for Baseline vs. 12 Months/Baseline vs. 24 Months; †p < 0.05 for 12 Months vs. 24 Months. Hodges‐Lehmann column shows paired median change (95% CI) from baseline to 24 months.

^a Friedman χ ² test (df = 2) on 366 paired observations present at all three visits; no missing values.

Sub‐group analysis showed a significant reduction in HbA1c over 24 months among study participants who were already on a basal–bolus regimen prior to enrolment and also those who were previously on other insulin regimens [median change of −20.8 mmol/mol (95% CI: −31.47 to −10.06) equivalent to −1.90% (95% CI: −2.88 to −0.92) for basal–bolus regimen of insulin treatment versus −21.3 mmol/mol (95% CI: −28.54 to −14.10) equivalent to −1.95% (95% CI: −2.61 to −1.29) for other regimens (p = 0.814)]. Details are given in Figure 2 and Table S1.

FIGURE 2.

Line diagram showing Median HbA1c of sub‐groups based on regimen of insulin treatment during the study period.

At baseline, 9.0% of study participants had an episode of diabetic ketoacidosis, and 56.0% had at least one episode of symptomatic or documented hypoglycaemia in the two months preceding enrolment. During the 24‐month study period, there were no episodes of diabetic ketoacidosis with no hospital admissions or deaths. After 12 months, the frequency of episodes of symptomatic or documented hypoglycaemia reduced to 47.8%, which further reduced to 40.9% after 24 months (p < 0.0001). All pairwise comparisons remained significant, with an absolute risk difference of 15.1% (95% CI: 7.9% to 22.3%) between baseline and 24 months.

Mean height increased from 116.9 ± 11.4 cm at baseline to 122.0 ± 10.6 cm at 12 months and to 127.6 ± 9.9 cm at 24 months (p < 0.0001 for overall change) in the age group < 10 years. In the age group 10–15 years, mean height increased from 141.0 ± 10.9 cm at baseline to 145.1 ± 10.2 cm at 12 months and to 149.7 ± 8.5 cm at 24 months (p < 0.0001 for overall change). Height remained constant in the age group > 15 years (p = 0.065). Mean weight increased from 20.1 ± 4.8 kg at baseline to 22.0 ± 4.8 kg at 12 months and to 25.6 ± 6.3 kg at 24 months (p < 0.0001 for overall change) in the age group < 10 years. In the age group 10–15 years, mean weight increased from 32.9 ± 8.1 kg at baseline to 36.2 ± 8.9 kg at 12 months and to 41.3 ± 10.1 kg at 24 months (p < 0.0001 for overall change). In the age group > 15 years, mean weight increased from 50.1 ± 9.9 kg at baseline to 51.0 ± 9.3 kg at 12 months and to 52.7 ± 10.6 kg at 24 months (p = 0.015 for overall change).

3.3. Evaluation of psychological well‐being of study participants and their caregivers

At baseline, all PedsQL™ instruments demonstrated high internal consistency (Cronbach's α 0.82–0.94) for both person self‐reports and parent proxy‐reports, with minimal floor effects (0.0%–0.5%) and ceiling effects ≤ 12.5%. This indicated good reliability and adequate score dispersion (Table 3.1).

TABLE 3.1.

Psychological well‐being of study participants and their caregivers: Baseline psychometric quality.

Summary scores	Items	Reporter	Cronbach α	% Floor	% Ceiling
Diabetes‐specific health‐related quality of life (Peds QL™ 3.2 Diabetes Module) a
Total scale score	33	Person	0.90	0.0	0.5
		Parent	0.92	0.0	0.0
Diabetes symptoms summary score	15	Person	0.82	0.0	0.5
		Parent	0.83	0.0	0.0
Diabetes management summary score	18	Person	0.88	0.0	3.8
		Parent	0.90	0.0	2.2
General well‐being (Peds QL™ 4.0 Generic Core Scales) b
Total scale score	23	Person	0.91	0.0	0.5
		Parent	0.89	0.0	1.1
Physical health summary score	8	Person	0.85	0.0	12.0
		Parent	0.85	0.0	8.7
Psychosocial health summary score	15	Person	0.87	0.0	0.5
		Parent	0.85	0.0	1.1
Family functioning and social support (Peds QL™ Family Impact Module Version 2.0) c
Total scale score	36	Parent	0.94	0.0	1.1
Parent HRQL summary score	20	Parent	0.90	0.0	2.2
Family functioning summary score	8	Parent	0.87	0.5	12.5

Note: Higher scores = better HRQoL; Floor/ceiling < 15% indicates adequate dispersion; Cronbach α ≥ 0.70 = acceptable internal consistency.

^a For the Peds QL™ 3.2 Diabetes Module, Diabetes Symptoms Summary Score = Diabetes Symptoms Scale Score; Diabetes Management Summary Score = Sum of the items over the number of items answered in the Treatment I, Treatment II, Worry and Communication Scales; Total Scale Score: Sum of all the items over the number of items answered on all the Scales.

^b For the Peds QL™ 4.0 Generic Core Scales, Physical Health Summary Score = Physical Functioning Scale Score; Psychosocial Health Summary Score = Sum of the items over the number of items answered in the Emotional, Social and School Functioning Scales; Total Scale Score = Sum of all the items over the number of items answered on all the Scales.

^c For the Peds QL™ Family Impact Module Version 2.0, Parent HRQL Summary Score = Sum of the items over the number of items answered in the Physical, Emotional, Social and Cognitive Functioning Scales; Family Functioning Summary Score = Sum of the items over the number of items answered in the Daily Activities and Family Relationships Scales; Total Scale Score = Sum of all the items over the number of items answered on all the Scales.

Over the 24‐month study period, there was steady and clinically meaningful improvement in diabetes‐specific HRQOL. Person self‐reported score increased by 5.5 points (95% CI 4.0–7.0), meeting MCID, with improvements reported in both diabetes symptoms (+4.9) and diabetes management (+6.1). General well‐being scores increased by 4.7 points (MCID met), mainly driven by improvement in psychosocial health summary score (+4.9) and school functioning (+5.4) (Table 3.2).

TABLE 3.2.

Psychological well‐being of study participants and their caregivers: Person self‐report (Results from linear mixed‐effects model (random intercept for participant; fixed effect = time: baseline, 12 months, 24 months)).

Dimensions	Baseline Mean ± SD (n = 366) a	12 Months Mean ± SD (n = 352) a	24 Months Mean ± SD (n = 349) a	Mean Change (95% CI) Baseline → 24 Months	MCID b	Global p#
Diabetes‐specific health‐related quality of life (Peds QL™ 3.2 Diabetes Module)
Person self‐report	70.7 ± 14.9	72.9 ± 15.2*	76.2 ± 15.4*†	+5.5 (95% CI: +4.0 to +7.0)	5.2	< 0.0001
Diabetes symptoms	67.2 ± 15.6	69.3 ± 15.9*	72.1 ± 17.1*†	+4.9 (95% CI: +3.3 to +6.5)	‐	< 0.0001
Diabetes symptoms summary score	67.2 ± 15.6	69.3 ± 15.9*	72.1 ± 17.1*†	+4.9 (95% CI: +3.3 to +6.5)	5.5	< 0.0001
Treatment barriers	70.8 ± 21.5	72.7 ± 21.7*	76.8 ± 21.8*†	+6.0 (95% CI: +3.7 to +8.3)	‐	< 0.0001
Treatment adherence	77.1 ± 19.9	79.3 ± 19.6*	82.8 ± 17.6*†	+5.7 (95% CI: +3.8 to +7.6)	‐	< 0.0001
Worry	60.4 ± 27.3	63.7 ± 27.3*	70.5 ± 27.2*†	+10.1 (95% CI: +7.2 to +13.0)	‐	< 0.0001
Communication	81.9 ± 20.4	83.8 ± 19.7*	85.6 ± 17.9*†	+3.6 (95% CI: +1.7 to +5.5)	‐	0.0002
Diabetes management summary score	73.6 ± 17.5	75.8 ± 17.5*	79.7 ± 17.4*†	+6.1 (95% CI: +4.3 to +7.8)	5.0	< 0.0001
General well‐being (Peds QL™ 4.0 Generic Core Scales)
Person self‐report	70.7 ± 17.8	72.2 ± 17.7*	75.5 ± 18.3*†	+4.7 (95% CI: +3.1 to +6.4)	4.3	< 0.0001
Physical functioning	73.9 ± 21.1	75.2 ± 20.8*	78.3 ± 20.5*†	+4.4 (95% CI: +2.4 to +6.3)	‐	< 0.0001
Physical health summary score	73.9 ± 21.1	75.2 ± 20.8*	78.3 ± 20.5*†	+4.4 (95% CI: +2.4 to +6.3)	6.6	< 0.0001
Emotional functioning	59.5 ± 22.9	61.3 ± 23.8*	64.2 ± 24.8*†	+4.7 (95% CI: +2.5 to +6.9)	‐	< 0.0001
Social functioning	81.0 ± 20.4	82.5 ± 20.2*	85.7 ± 18.2*†	+4.6 (95% CI: +2.9 to +6.3)	‐	< 0.0001
School functioning	66.7 ± 23.1	67.8 ± 22.2	72.1 ± 22.4*†	+5.4 (95% CI: +3.3 to +7.5)	‐	<0.0001
Psychosocial health summary score	69.1 ± 18.5	70.5 ± 18.5*	73.9 ± 18.9*†	+4.9 (95% CI: +3.3 to +6.6)	5.3	<0.0001

Note: Means ±95% CIs are estimated marginal means from a linear mixed‐effects model (random intercept for participant; fixed effect = time: baseline, 12 months, 24 months). Fourteen participants missed the 12‐month survey and 17 missed the 24‐month survey; their baseline HbA1c and PedsQL did not differ from completers (p > 0.05). Multiple testing: Three pairwise contrasts (baseline ↔ 12 months, baseline ↔ 24 months, 12 months ↔ 24 months) were Holm‐adjusted; significant contrasts are flagged *p < 0.05 for Baseline vs. 12 Months/Baseline vs. 24 Months; †p < 0.05 for 12 Months vs. 24 Months. Direction of scale: Higher scores = better diabetes‐specific HRQOL. Robustness check: Complete‐case Friedman/Wilcoxon results are reported in Supplementary Table S2.

^a Sample sizes for this table: baseline n = 366, 12 months n = 352, 24 months n = 349; paired complete‐case cohort n = 349. Column headings show the observed n per visit. Mixed model applied on available person at each visit.

^b Minimal clinically important differences (MCIDs) for the Peds QL™ 3.2 Diabetes Module and Peds QL™ 4.0 Generic Core Scales were based on published thresholds from Varni et al. (2018) and Varni et al. (2003), respectively, defined as approximately ≥ 5‐point change for total scale scores and summary scores.

There was also consistent improvement in parent proxy‐report of diabetes‐specific HRQOL which increased by 4.4 points (MCID met), with the diabetes management summary score (+5.1) exceeding its MCID. General well‐being improved by 6.3 points (MCID met), with notable increases in physical health summary score (+6.8) and psychosocial health summary score (+6.1). In the Family Impact Module, parents reported improvements across all domains, especially daily activities (+8.2) and worry (+7.5). All changes were statistically significant (p < 0.0001) (Table 3.3).

TABLE 3.3.

Psychological well‐being of study participants and their caregivers: Parent proxy‐report (results from linear mixed‐effects model (random intercept for participant; fixed effect = time: Baseline, 12 months, 24 months)).

Dimensions	Baseline Mean ± SD (n = 366) a	12 Months Mean ± SD (n = 352) a	24 Months Mean ± SD (n = 349) a	Mean Change (95% CI) Baseline → 24 Months	MCID b	Global p#
Diabetes‐specific health‐related quality of life (Peds QL™ 3.2 Diabetes Module)
Parent proxy‐report	66.3 ± 15.9	68.2 ± 16.1*	70.6 ± 16.3*†	+4.4 (95% CI: +3.3 to +5.5)	4.5	< 0.0001
Diabetes symptoms	64.6 ± 15.3	66.5 ± 16.6*	68.1 ± 17.1*†	+3.5 (95% CI: +2.3 to +4.8)	‐	< 0.0001
Diabetes symptoms summary score	64.6 ± 15.3	66.5 ± 16.6*	68.1 ± 17.1*†	+3.5 (95% CI: +2.3 to +4.8)	5.1	< 0.0001
Treatment barriers	67.1 ± 23.7	68.6 ± 23.5*	69.2 ± 24.8*	+2.1 (95% CI: +0.4 to +3.8)	‐	0.015
Treatment adherence	73.3 ± 22.9	75.9 ± 21.7*	79.3 ± 19.4*†	+6.0 (95% CI: +4.2 to +7.9)	‐	< 0.0001
Worry	52.3 ± 29.8	54.8 ± 28.7*	60.8 ± 27.5*†	+8.6 (95% CI: +6.0 to +11.1)	‐	< 0.0001
Communication	71.4 ± 24.1	72.7 ± 23.2	76.2 ± 22.4*†	+4.8 (95% CI: +2.9 to +6.6)	‐	< 0.0001
Diabetes management summary score	67.6 ± 20.2	69.7 ± 19.8*	72.7 ± 19.1*†	+5.1 (95% CI: +3.8 to +6.4)	5.0	< 0.0001
General well‐being (Peds QL™ 4.0 Generic Core Scales)
Parent proxy‐report	63.6 ± 18.5	65.7 ± 19.4*	69.9 ± 18.8*†	+6.3 (95% CI: +4.6 to +8.1)	4.5	< 0.0001
Physical functioning	63.2 ± 24.1	64.9 ± 24.6*	70.1 ± 23.7*†	+6.8 (95% CI: +4.5 to +9.1)	‐	< 0.0001
Physical health summary score	63.2 ± 24.1	64.9 ± 24.6*	70.1 ± 23.7*†	+6.8 (95% CI: +4.5 to +9.1)	6.9	< 0.0001
Emotional functioning	57.7 ± 20.9	60.6 ± 22.7*	64.4 ± 22.7*†	+6.7 (95% CI: +4.9 to +8.6)	‐	< 0.0001
Social functioning	73.0 ± 23.5	74.9 ± 22.9*	78.7 ± 21.4*†	+5.7 (95% CI: +3.4 to +8.0)	‐	< 0.0001
School functioning	60.9 ± 25.1	62.6 ± 24.6*	66.8 ± 23.1*†	+5.9 (95% CI: +3.4 to +8.4)	‐	< 0.0001
Psychosocial health summary score	63.9 ± 18.4	66.1 ± 19.1*	69.9 ± 18.3*†	+6.1 (95% CI: +4.3 to +7.9)	5.4	< 0.0001
Family functioning and social support (Peds QL™ Family Impact Module Version 2.0)
Parent proxy‐report	68.2 ± 17.0	69.8 ± 17.9*	73.7 ± 18.1*†	+5.5 (95% CI: +4.1 to +6.9)	‐	< 0.0001
Physical functioning	68.5 ± 21.5	69.7 ± 22.0*	73.6 ± 22.2*†	+5.1 (95% CI: +3.1 to +7.1)	‐	< 0.0001
Emotional functioning	59.1 ± 26.3	60.7 ± 26.8*	65.7 ± 25.6*†	+6.6 (95% CI: +4.5 to +8.8)	‐	< 0.0001
Social functioning	72.1 ± 24.0	73.8 ± 23.5*	76.4 ± 23.2*†	+4.3 (95% CI: +2.0 to +6.5)	‐	0.0002
Cognitive functioning	72.7 ± 22.4	74.5 ± 23.1*	77.1 ± 22.0*†	+4.5 (95% CI: +2.6 to +6.3)	‐	< 0.0001
Parent HRQL summary score	67.9 ± 18.2	69.5 ± 19.2*	73.1 ± 19.5*†	+5.2 (95% CI: +3.6 to +6.7)	‐	< 0.0001
Communication	74.1 ± 23.3	74.9 ± 23.5	77.9 ± 23.7*†	+3.8 (95% CI: +1.8 to +5.8)	‐	0.0002
Worry	58.6 ± 23.0	60.6 ± 24.1*	66.1 ± 23.7*†	+7.5 (95% CI: +5.4 to +9.7)	‐	< 0.0001
Daily activities	61.4 ± 28.3	63.4 ± 28.8*	69.6 ± 27.3*†	+8.2 (95% CI: +5.6 to +10.8)	‐	< 0.0001
Family relationships	79.8 ± 21.4	80.8 ± 21.4*	84.1 ± 19.3*†	+4.2 (95% CI: +2.7 to +5.8)	‐	< 0.0001
Family functioning summary score	72.9 ± 20.8	74.3 ± 21.3*	78.6 ± 19.7*†	+5.7 (95% CI: +4.0 to +7.4)	‐	< 0.0001

Note: Means ±95% CIs are estimated marginal means from a linear mixed‐effects model (random intercept for participant; fixed effect = time: baseline, 12 months, 24 months). Fourteen participants missed the 12‐month survey and 17 missed the 24‐month survey; their baseline HbA1c and PedsQL did not differ from completers (p > 0.05). Multiple testing: Three pairwise contrasts (baseline ↔ 12 months, baseline ↔ 24 months, 12 months ↔ 24 months) were Holm‐adjusted; significant contrasts are flagged *p < 0.05 for Baseline vs. 12 Months/Baseline vs. 24 Months; †p < 0.05 for 12 Months vs. 24 Months. Direction of scale: Higher scores = better diabetes‐specific HRQOL. Robustness check: Complete‐case Friedman/Wilcoxon results are reported in Table S2.

^a Sample sizes for this table: baseline n = 366, 12 months n = 352, 24 months n = 349; paired complete‐case cohort n = 349. Column headings show the observed n per visit. Mixed model applied on available person at each visit.

^b Minimal clinically important differences (MCIDs) for the Peds QL™ 3.2 Diabetes Module and Peds QL™ 4.0 Generic Core Scales were based on published thresholds from Varni et al. (2018) and Varni et al. (2003), respectively, defined as approximately ≥ 5‐point change for total scale scores and summary scores. The Peds QL™ Family Impact Module Version 2.0 does not have specific MCID scores. The Family Impact Module assesses how a participant's health condition affects the family, and scores explore overall impact on family functioning.

Results from the complete‐case analysis were found to be consistent with main results, with similar directions and magnitudes of improvements across all domains, confirming robustness of results. Small differences in estimates were observed, likely due to reduced sample size (Table S2).

3.4. Evaluation of direct and indirect expenditure incurred by families

There was a reduction in total median expenditure incurred by families per month from 2600 INR (1730–4100) at baseline to 200 INR (100–550) at 24 months of implementation of the model (p < 0.0001), with a median reduction of −2690 INR (95% CI: −3217 to −2163). Median direct monthly expenditure incurred by families fell sharply from 2200 INR (1200–3615) to 0 (p < 0.0001), with falls in expenditure incurred for insulin (p < 0.0001), blood glucose measuring devices/blood glucose test strips (p < 0.0001), insulin syringes (p = 0.001) and laboratory investigations (p = 0.009). There was also a trend to reduction in median indirect expenditure incurred by families per month (p = 0.053). Details of direct and indirect expenditure incurred by families per month are shown in Table 4.

TABLE 4.

Direct and indirect expenditure incurred per month, INR ^a .

Variables	Baseline Median (IQR) (n = 366)	24 Months Median (IQR) (n = 349)	Wilcoxon Signed‐Rank Test (p)	Hodges‐Lehmann Median Change (95% CI) Baseline → 24 Months
Direct expenditure incurred by families per month, INR a	2200 (1200–3615)	0 (0–0)	< 0.0001	−1760 (95% CI: −2434 to −1086)
Insulin	500 (0–1050)	0 (0–0)	< 0.0001	−300 (95% CI: −683 to +82.60)
Blood glucose measuring devices/ blood glucose test strips	900 (500–1200)	0 (0–0)	< 0.0001	−900 (95% CI: −1035 to −765)
Insulin syringes	90 (0–240)	0 (0–0)	0.001	−80 (95% CI: −148 to −12.50)
Laboratory investigations	0 (0–550)	0 (0–0)	0.009	–
Indirect expenditure incurred by families per month, INR a	500 (230–870)	200 (92.50–538)	0.053	−205 (95% CI: −421 to +11.30)
Transport to health facility	200 (150–500)	150 (75–200)	0.194	−45 (95% CI: −101.30 to +11.30)
Food	100 (0–200)	0 (0–50)	0.505	−10 (95% CI: −57.50 to +37.50)
Loss of wages	0 (0–300)	0 (0–338)	0.306	0 (95% CI: −131 to +131)
Total expenditure incurred by families per month, INR a	2600 (1730–4100)	200 (100–550)	< 0.0001	−2690 (95% CI: −3217 to −2163)

^a 1 US$ = INR 87.62 Reserve Bank of India. ²⁰

Details of cost of model to the health care system/ funding allocation are given in Table S3.

4. DISCUSSION

This study reports significant improvement in clinical and psychological well‐being outcomes 12 months and 24 months after introduction of the West Bengal Model for T1D care in individuals living with T1D and their parents/caregivers. This includes improvements in glycaemic control as demonstrated by reduction in median HbA1c from 79 mmol/mol (9.4%) to 64 mmol/mol (8.0%), no episodes of diabetic ketoacidosis, no hospitalizations, no deaths and reduced episodes of hypoglycaemia, healthier insulin injection sites, improved growth and development and psychological well‐being, with person self‐reported HRQOL increasing by 5.5 points and meeting the MCID, and reduced family T1D‐related expenditure from 2600 INR to 200 INR per month.

Although international clinical guidelines recommend structured T1D care models for facilitating best person outcomes, ³ , ⁹ such care is available in very few centres/care facilities in India, which are mostly private or supported by other organizations. International guidelines recommend basal–bolus insulin regimens as standard of care for T1D, with use of premixed insulin not recommended. ³³ In India, however, the most common standard of care is premixed insulin with or without SMBG. This ‘minimal care’ level is associated with substantially poorer outcomes and reduced life expectancy. ³ , ⁶ , ³⁴ The single‐centre study by Madhu et al. reported that 11.6% of persons with T1D died over a mean follow‐up of just 6.4 years, with the majority of the study group having inadequate glycaemic control [HbA1c > 86 mmol/mol (10%)]. Most of the deaths were from preventable causes such as diabetic ketoacidosis, infections and renal failure. The study revealed gaps in diagnosis, treatment adherence and follow‐up in a tertiary care setting. ⁷ Considering these poor clinical outcomes and high mortality rate among individuals living with T1D in India, ⁶ , ⁷ , ⁸ our Institutional Ethics Committee dissuaded us from having the present study being a randomized controlled trial.

Prior to implementation of this model, the majority of study participants received unstructured care from endocrinologists at various government/non‐government health care facilities (predominantly government tertiary health care facilities) and were being treated with premixed insulin with or without SMBG. Significant improvements in a range of clinical outcomes were achieved within 24 months despite a high proportion of study participants being of lower socio‐economic status and the majority of their parents/caregivers having not completed primary education. As described by Ogle et al. ³ expected outcomes of ‘intermediate care’ include mean HbA1c levels between 64 and 80 mmol/mol (8.0%–9.5%) and infrequent mortality. These outcomes were achieved in the present study. It also exceeds the MCID of 4 mmol/mol (0.4%) established in previous literature. ¹⁵ , ¹⁶ It is notable that similar improvements in HbA1c were observed across all participants, including those already on a basal–bolus regimen at enrolment and those who transitioned from other insulin regimens, underscoring the comprehensive effectiveness of the model. This study provides convincing evidence that the model is associated with improved outcomes, better glycaemic control and well‐being.

These positive study results were achieved despite care being delivered by physicians/paediatricians (non‐specialist teams), who were trained at secondary level and mentored by qualified endocrinologists. Team delivering care also included nursing staff and NCD counsellors who were given basic training to deliver diabetes‐related nutritional advice and education. Upskilling of personnel was accompanied by maintenance of a steady insulin supply chain, monitoring and mentoring, continuous auditing of processes and outcomes, and real‐time electronic and physical prescription and data keeping.

The model was piggybacked on the existing adult NCD program in India, which mandates that every district hospital should have an adult NCD clinic. Such clinics include physical space with desks, chairs, weighing machine, stadiometer, sphygmomanometer, computer with internet services, and human resources of clinician, nursing staff, district NCD counsellor and data entry operator. In this study, these health professionals were trained and supported, and staff rotation (being a common problem in government hospitals) was avoided.

The model was designed for long‐term sustainability by piggybacking on the existing district hospital infrastructure at secondary level of healthcare without recruitment of additional human resources, upgrading existing NCD clinics and identifying and training general physicians, nurses and NCD counsellors rather than specialists. In addition to leveraging existing NCD health service delivery infrastructure, an essential component was financial support provided by both state and national governments. The model is funded by the Government of West Bengal and part‐funded by the Central Government, ensuring continuity beyond the study period. Based on the success of this pilot phase, the model has been scaled up to ten more health districts of West Bengal (15 districts in total) and now caters to 1500 T1D subjects. In India, some organizations such as Changing Diabetes in Children, ³⁵ Life for a Child ¹¹ and several other non‐governmental organizations support enhancements of T1D care. However, these initiatives lack state‐wide, let alone country‐wide coverage and are therefore limited in their impact and long‐term sustainability.

By providing this care at district hospitals, children with T1D and their families were able to access care in their home districts rather than the need to travel long distances to access tertiary health care facilities that are only available in the state capital of Kolkata.

While the study does not directly address causal comparisons of different T1D care models, it provides insight into implementing and delivering structured, decentralized T1D care in LMIC settings where baseline services were limited. By integrating within the implementation science framework, the study highlighted how human resource allocation, training, mentoring and steady supply chains contribute to sustainable care delivery. These positive study results provide a template, which can be scaled up to other district hospitals of West Bengal and elsewhere in India (with government support and funding) with modification/adaptation as required. This model could potentially be adopted in other low‐ and middle‐income countries.

Future research will investigate phased additions to the current model of care, including specialized health care professionals such as dedicated diabetes educators, dietitians, and clinical psychologists, along with the implementation of intensive diabetes education, regular diabetes educational camps and incorporation of support groups. In addition, community engagement and awareness building via the inclusion of T1D in the Accredited Social Health Activist's (ASHA) Community‐Based Assessment Checklist (CBAC) Form will improve early diagnosis and referral pathways. These elements were purposefully excluded in the first phase of our model to demonstrate the effectiveness of each of the additional components. This is vital as each component has associated financial implications and requires evaluation of its costs and benefits to justify government support of additional resources.

Strengths of this study include that all participants completed the 24‐month follow‐up, with all adhering to a basal–bolus insulin regimen and a mean of two SMBGs a day. Electronic data capture, with regular clinic level auditing, enabled robust and complete data to be collected in a harmonized and systematic manner across five different clinics. In addition, the central provision of health care provider training, support and education means that the model of care was implemented uniformly across and within the centres.

Despite providing compelling evidence for the model's effectiveness, the study has several limitations. The absence of a control group due to ethical concerns surrounding high mortality rates in unstructured T1D care settings in India limits the ability for causal inference regarding the effectiveness of the intervention and to control for external confounding factors or natural disease progression. Moreover, although ethically justified, the absence of a randomized control group makes the study susceptible to biases such as the Hawthorne effect, regression to the mean, secular trends in diabetes care or maturation effects over time. A randomized or stepped‐wedge design could not be implemented because the intervention was delivered as a government‐funded public health model, and withholding or delaying structured care would be unethical in this setting. The study was limited to five districts in one state, which may limit the direct generalizability of outcomes to other settings with different healthcare infrastructure or socio‐economic background without further adaptation.

5. CONCLUSION

In conclusion, implementation of the West Bengal Model for T1D Care (K1DS), a dedicated government‐funded structured model of care emphasizing training, mentoring and consistent supply chains, resulted in improved outcomes for persons and parent/caregivers in secondary‐level (district) hospitals. This scalable model provides a template for implementation across the state of West Bengal and other parts of India and can inform future national health policy developments aimed at improving care of children with T1D.

AUTHOR CONTRIBUTIONS

S.G. conceptualized the model and study, imparted training, procured funds, aligned administration and stakeholders and reviewed the draft manuscript. M.Y. was responsible for implementation of the model, data collection, auditing, data analysis and wrote first draft of the manuscript. P.M. analysed the data and reviewed the manuscript. B.P. helped in writing proposal document of the model. D.K.M. helped in questionnaire selection. G.D.O. and P.K. provided advice and guidance and helped editing the manuscript. All authors are the guarantors of this work and, as such, had full access to all the data in the article and take responsibility for the integrity of the data and the accuracy of the data analysis.

FUNDING

We gratefully acknowledge the entire administration of The Government of West Bengal, Department of Health and Family Welfare for supporting and funding the model of T1D care through state National Health Mission under activity heads monitoring, miscellaneous and contingency (vide Memo No: HFW‐27024/19/2020‐NCD SEC‐Dept. of H&FW/369/2021, dated: 08.07.2021) and strengthening of laboratory, for procurement of biomedical equipment (vide Memo No: HFW‐27024/13/2022‐NHM SEC‐ Dept. of H&FW/664/2022, dated: 23.06.2022). We are also grateful to The Government of India for providing financial support through program implementation plan (PIP). We acknowledge University Grants Commission for providing fellowship (NTA Ref. No.: 200510265896). We acknowledge United Nations Children’s Fund (UNICEF) and Diabetes & Endocrine Research Trust (DERT) for their partnership since December 2024.

CONFLICT OF INTEREST STATEMENT

The authors declare no conflicts of interest.

Supporting information

Table S1. Sub‐group analysis of outcomes based on regimen of insulin treatment.

Table S2. Sensitivity analysis of health‐related quality of life scores using complete‐case cohort.

Table S3. Details of cost to the health care system/funding allocation.