Original Research••

How do sex-specific BMI trajectories shape diabetes risk? A longitudinal analysis of Indonesian adults

Abstract

Introduction Sex is a critical predictor of body mass index (BMI) trajectory over the life span, playing a role in disparities in the risk of diabetes. While there is some evidence of the effect of BMI on the risk of diabetes, little is known about sex differences in BMI trajectories over the life span and their later life association with diabetes, especially in low-income and middle-income settings.

Methods Using panel data from the Indonesian Family Life Survey, this study examines the impact of an individual’s BMI trajectory throughout adulthood on diabetes onset. Analysis for men and women were conducted separately. First, growth curve modelling estimates individuals’ BMI trajectories over age. Second, the effect of BMI trajectories on diabetes is estimated using logistic regression adjusted for individual socioeconomic status. Finally, we perform relative dominance analysis to test the importance of BMI trajectories as a predictor of diabetes in later life against abdominal obesity measurements.

Results BMI trajectories over adulthood strongly predict the probability of diabetes in both men and women. A rapidly increasing BMI significantly increases the probability of diabetes in men regardless of individuals’ initial BMI. Among women, those who started out with an overweight/obese BMI and experienced a rapid loss in BMI over age had the highest risk of diabetes. Greater educational attainment is associated with an increase in the likelihood of diabetes in men, but higher education level is a protective factor from diabetes in women.

Conclusion The findings suggest that maintaining BMI at normal cut-off over the life course can lower the likelihood of diabetes onset in later adulthood. This study emphasises that simple monitoring of BMI trajectory over adulthood could be a useful tool to identify the population at risk of diabetes in contexts with substantial underdiagnoses of diabetes.

What is already known on this topic

  • Little research has focused on sex differences in role of body mass index (BMI) trajectories on diabetes diagnosis, particularly in low-income and middle-income settings.

What this study adds

  • This study conducts sex-stratified analysis to estimate the effect of BMI trajectories on diabetes in later life in a low-middle income setting that is experiencing rapid increases in obesity, has a high proportion of underdiagnosed diabetes and is rapidly ageing.

  • Taking advantage of longitudinal measurements of BMI over 3–5 time periods, this study identifies the role of individual BMI trajectory over age in impacting diabetes in later life as identified by glycosylated haemoglobin measurement.

  • We identified that rapid decline in BMI over age is linked to a higher risk of diabetes among women suggesting long-term effect of undiagnosed diabetes, but it is not found for men. This is essential information to identify populations at risk specifically in low-income and middle-income countries that have a high burden of undiagnosed diabetes.

How this study might affect research, practice or policy

  • This study highlights the need for consistent monitoring of BMI over adulthood, and sex-based prevention and screening programmes to tackle the global increase of diabetes.

Introduction

Type 2 diabetes (hereafter diabetes) presents an urgent challenge to global health and has rapidly risen to the third-leading risk factor for mortality and morbidity worldwide.1 Onset of diabetes is linked to individual-level risk factors including dietary behaviours and physical activity, as well as structural and societal determinants including availability and affordability of fresh food, a move towards sedentary, service-based occupations and increasing urbanisation.2 Although commonly thought of as a problem in more affluent countries, more than 80% of adults with diabetes currently live in low-income or middle-income countries (LMICs).3 4 Public health systems in LMICs are struggling to adapt to this growing burden of diabetes, and recent evidence suggests that screening, linkage to care, treatment and blood sugar control are all very limited.5–8

Elevated body mass index (BMI) is a well-established contributor to diabetes risk. Prior research has found that rapid increases in weight over age, early age of obesity onset and a longer duration in obesity are among the largest predictors of diabetes onset.9–11 Published literature investigating the link between long-term BMI trajectories and diabetes onset suggests that both the pace of BMI change and individual-level factors such as age and sex act to meaningfully shape diabetes risk.9 12–14 However, a recent systematic review on group based BMI trajectory modelling identified an inconsistent association between declining in BMI over adulthood and the risk of diabetes in two different populations, highlighting the need for further study.11 Women may be at greater risk of developing diabetes following weight gain due to differences in hormone levels, insulin sensitivity and fat deposition.15–17 However, prior evidence on BMI trajectories and diabetes onset has often relied on single-sex surveys, and even studies including both men and women have rarely explored how sex shapes BMI trajectories in ways that may differentially impact diabetes risk.11 To our knowledge, only one study has investigated sex-stratified BMI trajectories as a predictor of diabetes onset in adults. This study found substantial heterogeneity by sex in connection between BMI trajectories and diabetes incidence.11 18 In addition, the association between BMI and diabetes is known to differ by geography and ethnicity.19 20

Understanding the links between BMI change and diabetes onset is particularly important for populations in Asia. Previous evidence has shown that individuals in South and East Asia may be at risk of diabetes even at lower levels of BMI as compared with individuals of European or African descent.21 22 This elevated risk of diabetes is especially concerning given the recent increases in the prevalence of overweight and obesity experienced in the region. By 2030, nearly two-thirds of adults in South and Southeast Asia are projected to fall into either overweight or obese BMI classifications.23 At the same time, countries throughout the region are also undergoing a rapid process of population ageing.24 Combined, these rising trends in overweight/obesity and population ageing pose a substantial threat to population health in the region.

Among LMICs, Indonesia faces one of the greatest challenges in managing its burden of diabetes.7 25 Indonesia is the fourth-largest country in the world by population and is experiencing rapid population ageing, with the population aged 60 and over projected to grow from 10.7% of the population in 2020 to 19.9% by 2045.26 Indonesia is among the top ten countries with the largest number of people living with diabetes. The prevalence of diabetes in adults aged 20–79 was 10.7% in 2019 and is expected to reach 16.6% by 2045.25 The country was estimated to have the world’s third-highest burden of undiagnosed diabetes in adults aged 20–70 years, with estimated 73.7% (14.3 million) of person with diabetes lacking a formal diagnosis in 2021.25 This rise in diabetes prevalence is a major contributor to mortality in Indonesia and was the fourth-leading cause of years of life lost in 2019.27 Recent evidence suggests that the healthcare system is struggling to manage a growing burden of non-communicable diseases; less than a quarter of individuals with diabetes were receiving treatment, and healthcare provider knowledge about diabetes and diabetes management was generally poor.7 28 Concurrently, the country is among the top 10 countries with the most rapid rise of overweight/obesity, which also contributes to increasing the population at risk of developing diabetes.29 This rapid rise in obesity affects population groups differently, and previous research has found disparities in BMI trajectory over the life span by sex and socioeconomic status (SES), meaning that there are likely to be substantial inequalities in the risk of diabetes across social groups.30

In this study, we evaluate the relationship between long-term BMI change and diabetes diagnosis among adults in Indonesia, and explore sociodemographic differences in this relationship. To do so, we take advantage of data from five waves (1993–2014) of the Indonesian Family Life Survey (IFLS) that measures individual’s weight and height and conducted a blood test for diabetes diagnoses. Using two stages of analysis comprising growth curve modelling and logistic regression, we examine how the pace of BMI change over age relates to a biomarker-assessed diabetes diagnosis based on glycosylated haemoglobin (HbA1c) values from a dried blood test. Our analyses explore two primary questions. First, how do the initial level and slope of an individual’s BMI trajectory impact their probability of having diabetes? And second, how do these relationships vary by sex and other sociodemographic characteristics? Combined, these analyses represent the first exploration of the impacts of long-term BMI change on diabetes in Indonesia, and provide unique insight into the role of sex-related and age-related changes in BMI on diabetes risk in a rapidly changing population.

Method and statistical analysis

Data source

This study uses data from the longitudinal IFLS. The survey follows households (HHs) and individuals through five surveys from 1993 to 2014. IFLS gathers information on individual-level health measurements including individual’s weight and height. In 2014, the survey conducted dried blood spot testing of HbA1c level, which is used for type 2 diabetes diagnosis. The survey also collects detailed sociodemographic information of individuals, including education status, employment, relationship status and residential moves. The survey also longitudinally collected information on HH assets, housing conditions, access to electricity and clean water, that are widely used as proxy measures of HH wealth. The IFLS datasets are publicly available and can be accessed through https://www.rand.org/well-being/social-and-behavioral-policy/data/FLS/IFLS/access.html.

We limit our analysis to individuals with at least three repeated measures of weight and height over the study period to increase the robustness of individuals BMI growth parameters (initial BMI and the pace of change). In addition, only individuals who participated in the 2014 wave of IFLS and have an HbA1c measurement are included in the analysis. In total, this study uses data from 1522 men and 2006 women. Flow chart for sample selection and a descriptive comparison between observations based on inclusion criteria are illustrated in online supplemental figure S1 and table S1.

Variables

Distal outcome: diabetes status

This study uses HbA1c level from a dried blood spot to classify diabetes cases. HbA1c is a measure of glucose metabolism used for diagnosis of diabetes and monitoring blood sugar level among people with diabetes. HbA1c level indicates the average of glucose in the blood over the past 2–3 months.31 Hence, the HbA1c test provides a reliable diagnosis of diabetes status and is recommended as a diagnostic criterion over the fasting glucose test, which only measures blood sugar at a single point in time.31 This study follows American Diabetes Association recommendation of an HbA1c cut-off of ≥6.5% for diabetes diagnosis.32

Main covariates of interest

The main covariate in this study is BMI trajectory, as represented by the initial BMI and pace of BMI change over age. The parameters are estimated from a growth curve model estimating BMI trajectory as a function of individual’s BMI over age (detailed on the Statistical analysis section).

Control covariates

Age and abdominal obesity

We use the age of each individual at last interview, which refers to age at which diabetes status was measured. Age is grouped into four categories: 30–39, 40–49, 50–59 and ≥60.

In sensitivity analyses, we also include abdominal obesity measures to evaluate the relative importance of this measure as compared with individual BMI trajectories. Abdominal obesity is widely known as a strong predictor of diabetes.33 34 However, it is not commonly collected in clinical settings, in contrast to BMI which is based on simple, measures of height and weight that are commonly collected in most interactions with the healthcare system. Hence, we additionally sought to investigate how the predictive power of BMI trajectory compared with information from a more refined measure of abdominal obesity. Abdominal obesity is determined from waist-to-hip ratio calculated from waist circumference measurement divided by hip measurement. Following the WHO guideline,35 the waist-to-hip ratio is classified into obese if the ratio ≥0.90 for men and ≥0.85 for women. As the waist circumference were measured only in individuals aged 40 and over, the sensitivity analysis is only conducted in this age group.

Socioeconomic variables

This study includes controls for SES to adjust for potential differences in the effect of individuals’ BMI trajectory on the probability of diabetes. Education level, employment status (unemployed, working and retired), type of occupation (non-manual, manual and farming) and relative HH wealth index are included in the analysis as measures of individual SES.

Education level refers to the highest level of education completed by the individual as of the last interview. Education is categorised into three groups: (1) no education/not completed primary school, (2) minimum primary school and (3) minimum secondary school. In this analysis, we only use employment status and type of jobs at the last interview. However, only education level and relative HH wealth are included in the final model because including employment status and type of occupation did not significantly improve the model.

Relative HH wealth index is constructed from HH-level data, consisting of information on (a) HH assets, (b) housing materials, (c) main of source drinking water, (d) toilet ownership and (e) electricity. The first stage of constructing the relative HH wealth index is to compute the cross-sectional composite measures of HH wealth index for each survey wave, stratified by urban and rural areas. Stratifying the wealth index by area and survey period was done to take account of variability of the value of HH assets over time, and of differences in components of wealth between urban and rural areas.36 Polychoric principal component analysis was applied to construct the wealth index as the index includes both dummy variables and ordinal scales. Relative HH wealth index is the mean of HH position in wealth quantile over time. Details on the construction of the relative wealth index can be found in online supplemental tables S2 and S3.

Statistical analysis

Our analyses are composed of a three-step process. First, we model individual BMI trajectories over age to estimate initial BMI level and pace of BMI change of an individual’s BMI trajectory. As BMI trajectories are known to be differentially patterned by sex over age,30 growth curve models are performed separately for men and women. The growth model estimates individual changes in BMI as a function of age and the quadratic of age. The initial BMI term is the predicted value of an individual’s relative intercept of BMI taking account the fixed and random components of the model. The pace of BMI change is the predicted value of individual’s relative slope/change of BMI over age. The initial BMI is categorised into three categories due to sample size limitations for obese observations. The categories are underweight (<18.5  Inline Formula ), normal  Inline Formula  and overweight/obese ( Inline Formula ), following recommendations for BMI cut-offs for Asian-Pacific population.37 Based on the distribution of the predicted value of individual’s relative slope, the pace of BMI change is classified into five quantiles: rapid loss, moderate loss, stable, moderate increase and rapid increase. Positive pace of BMI change refers to an increase in BMI over time and negative pace of change refers to loss of weight over time.

At the second step, these estimated initial BMI and pace of BMI change terms are used as input for analysis where we explore the impact of BMI trajectories on a diabetes diagnosis in later life. We use logistic regression to estimate the effect of BMI trajectories on a later life diabetes diagnosis, adjusted by individual SES. The effect of BMI trajectories is adjusted by including control variables that are age at diabetes diagnosis, SES (educational level, employment status, type of job, relative HH wealth index) and residential area.

Finally, we perform a dominance analysis to evaluate the importance of individual BMI trajectories as predictor for occurrence of diabetes in later life. This analysis calculates the reduction in prediction error that is correlated with each independent variable in the selected model.38 All analyses are performed using STATA V. 17.0.

Results

Estimated BMI growth parameters from growth curve modelling

In the first stage of analysis, we estimate BMI growth parameters (initial BMI and pace of change) of individuals using growth curve modelling, separately for men and women. The estimates BMI growth parameters and their distributions can be found in online supplemental table S4, online supplementalfigures S2 and S3). The predicted mean of initial BMI for women is 19.21  Inline Formula  with an SD of 2.88. The estimated mean of the pace of BMI change is 0.23 points of BMI increase for each year of age, with an SD of 0.09. For men, the mean initial BMI is 19.09  Inline Formula  with an SD of 2.11. The predicted mean of the pace of BMI change is 0.16 with an SD of 0.08. Based on their estimated growth parameters, observations are grouped into three categories of initial BMI (underweight, normal and overweight/obese) and five categories of pace of change (rapid loss, moderate loss, stable, moderate increase and rapid increase) as described above. These growth parameters are used as explanatory variables in subsequent analyses predicting the probability of diabetes in later life.

Table 1 presents the proportion individuals with HbA1c diagnosed with or without diabetes by BMI growth trajectory and sociodemographic covariates, separately for men and women. The table also presents χ2 results testing for differences between the group living with diabetes and the non-diabetic population. For women, the distribution of diabetes is significantly different by categories of initial BMI but does not significantly vary by the pace of change. However, for men, the diabetes distribution is significantly different for both BMI growth parameters. The distribution of diabetes cases significantly varies by age group among both men and women, with an increasing trend over age. Diabetes cases are not significantly different across SES groups in women, but vary significantly among men by education level and relative HH wealth. For both men and women, diabetes rates are higher among those living in urban areas as compared with rural areas. Diabetes is also more common among individuals with elevated waist-to-hip ratio.

Table 1
|
Sociodemographic characteristics of individuals based on HbA1c-diagnosed diabetes and p value of χ2 comparison to those without diabetes by sex, IFLS 2014

Impact of BMI trajectories on the probability of diabetes.

The second stage of analysis uses logistic regression to explore the role of BMI growth parameters and SES characteristics in shaping the risk of a diabetes diagnosis in later life. We find that both initial BMI and pace of BMI change are the important predictors of a diabetes diagnosis. The probability of having diabetes is highest in women with an initial BMI in the overweight/obese group. However, we find a significant interaction between the BMI growth parameters, suggesting that the effect of BMI trajectory over age may change depending on the level of initial BMI among women (table 2, column I). Overall, for women starting with an underweight or normal BMI, an increase in BMI over age elevates the risk of diabetes. However, for women with an overweight/obese initial BMI, the decline in BMI over age increases the risk of diabetes. The age-specific predicted probabilities of diabetes based on these interaction effects are illustrated in figure 1.

Table 2
|
ORs of the probability of a diabetes diagnosis based on BMI trajectory and SES, men and women with 95% CI
Figure 1
Request permissionrequest permission icon
Figure 1

Predicted probability of diabetes diagnosis by initial BMI group and pace of BMI change over age, women with 95% CI

As rapid loss in BMI may relate to both the severity of diabetes morbidity among people living with diabetes or to the ageing process among elderly individuals, we conduct a robustness test on the significant effect of the interaction BMI growth parameters by excluding individuals with historical diabetes (diagnosed by health practitioners) and individuals aged ≥60. The sensitivity analysis shows a consistent, significant effect in the interaction between initial BMI and the pace of BMI change over age among women (online supplemental table S5). This finding that the highest risk of diabetes is for overweight/obese women who experience a rapid loss in BMI persists among people aged <60 who have not previously received a diabetes diagnosis. Further data exploration (online supplemental tables S6 and S7) suggests that the rapid loss in BMI among overweight/obese women likely indicates morbidity related to diabetes. The proportion of diabetes cases among overweight/obese women experiencing a loss in BMI over time is 49.2% (29 out of 59 observations). This comprises 27.1% (16 cases out of 59) women ever diagnosed with diabetes and 22% (or 13 cases out of 59 observations) of individuals with undiagnosed diabetes. Taken together, this means that almost a quarter of overweight/obese women experiencing loss in BMI are unaware that they are living with diabetes at the time that the dried blood test was conducted. Rapid loss in BMI among overweight/obese women was more commonly found among women aged ≥50 and less educated (statistically significant at 10% level).

For men, initial BMI and pace of BMI change over age independently predict the probability of diabetes. The probability is significantly higher among individuals with an initial BMI of normal or overweight/obese compared with men starting at underweight (table 2, column III). Men who experience a rapid increase in BMI over age have a significantly higher likelihood of having diabetes in later life as compared with other groups. Figure 2 graphically presents the predicted probability of diabetes by groups of initial BMI and pace of BMI change over age for men. The interaction between two growth parameters of BMI is not significant in the men’s model, hence it is excluded from the final model (online supplemental table S8).

Figure 2
Request permissionrequest permission icon
Figure 2

Predicted probability of diabetes by age and initial BMI group and pace of BMI change over age, men with 95% CI

In the sensitivity analysis, the significant effect of initial BMI and pace of BMI change is consistent after adjustment of the waist-to-hip ratio measurement on the last interview for both men’s and women’s models (table 2, column II and IV). This suggests that the BMI growth parameters (initial BMI and pace of BMI change) are sufficient predictors of diabetes in the absence of information on abdominal obesity. This sensitivity analysis is only conducted for a subsample of those aged ≥40, as the IFLS only conducted waist circumference measures in this age group.

Keeping other covariates at observed values, the probability of diabetes increases substantially over age. The probability of diabetes is significantly higher among women at age 40 onwards, while for men diabetes risk rises significantly at age 50 and onwards (online supplemental figure S4 for illustration). This suggests that the risk of diabetes onset occurs at an earlier age among women as compared with men.

Further analysis on the role of SES shows that higher education level is protective to the risk of diabetes among women, as indicated by a significantly lower probability of diabetes among educated women compared with non-educated. In contrast, the risk of diabetes increases by education level among men. Relative HH index is a significant predictor for men only, with u-shape relationship with the probability of diabetes. The results show no significant differences in the probability of diabetes between urban and rural areas for both sexes.

Considering the high proportion of undiagnosed diabetes in Indonesia, we test the dominance of BMI growth parameters against other covariates in the final model and adjusted waist-to-hip ratio model of table 2. For women, information on initial BMI is the most important predictor, explaining 70.4% of variation in the probability of diabetes (table 3). Even after adjusting for waist-to-hip ratio, the importance of initial BMI is consistent, still ranking as the most important predictor with a slightly lower standardised dominance statistic of 65.3%. This suggests that BMI trajectories are the most important predictors of diabetes in later life for women, even in the absence of information on central obesity. The relative dominance analysis for men shows that age, initial BMI and pace of BMI change are almost equally important predictors of diabetes in men. Initial BMI accounts for around 28.6% of variation, and pace of BMI change explains 24.9% of variation in the probability of diabetes among men.

Table 3
|
Summary table of relative dominance analysis for men and women

Discussion

In a longitudinal study based in Indonesia, we found a number of novel insights into the relationship between long-term changes in BMI and diabetes diagnosis in later life. Increasing trajectory of BMI over the life course was strongly associated with diabetes diagnosis in later life among both men and women. For women, rapid declines in BMI in later adulthood among those starting with an overweight or obese BMI were also strongly associated with diabetes diagnosis in later life.

Some of the most noteworthy findings of our analyses relate to sex differences in the relationship between long-term BMI change and diabetes onset. We find that women seem to be vulnerable to diabetes at earlier ages as compared with men, suggesting that the age threshold for diabetes screening may need to be lower for women as compared with men in this setting. Socioeconomic factors also appear to have a differing influence on men’s and women’s risk of developing diabetes. Greater educational attainment was associated with a lower likelihood of having diabetes among women, but a substantially higher risk of diabetes among men. Similarly, HH wealth has little relationship with diabetes risk for women, but has a u-shaped relationship among men, with both lower and higher-wealth individuals at increased risk of diabetes. These sex-based differences in BMI trajectories, and their relationship with diabetes onset, highlight the need for more in-depth gender-based research to understand the mechanisms underlying these findings.

A relative dominance analysis found that measured BMI in early adulthood, and the pace of change in BMI over the life course, were the strongest predictors of a later-life diabetes diagnosis. Adjusting for waist-to-hip ratio, the analysis highlights the dominance of initial BMI as the most important predictor for women. Further, the standardised values suggest that initial BMI, waist-to-hip ratio measure, pace of BMI change and age are all strong predictors of diabetes for men. Both relative analysis for women and men demonstrate that the inclusion of waist-to-hip ratio reduces the importance of age in predicting diabetes. This indicates that the effect of age is less important compared with anthropometric measurements for predicting diabetes in women, but it is equally as important as anthropometric measurements for predicting diabetes in men.

Our analysis of the consequences of BMI trajectories over age on eventual diabetes diagnosis found that overweight/obese women who experienced a rapid decline in BMI had the highest risk of having diabetes in later life. Our findings partially align with a previous study in Austria that found the risk of diabetes was highest among men and women over the age of 50 with BMI trajectories starting with a plateau and then declining over time.18 In a context with very high underdiagnoses of diabetes, we theorise that long-term undiagnosed diabetes may be the cause, rather than the consequence, of this declining trajectory of BMI. Previous literature has shown that unintentional weight loss could be related to underlying or undiagnosed diseases, including undiagnosed and uncontrolled diabetes.39 Individuals with a longer duration in diabetes have previously been shown to experience a more rapid loss of muscle mass40 that indicates the severity of diabetes morbidity. Fluid loss due to excessive urination and resistance to insulin resulting in a decreased stimulation of protein synthesis pathways are two of the potential mechanisms that explain weight and muscle loss among people with diagnosed and undiagnosed diabetes.40 41 In particular, older women with diabetes have been shown to have a substantially higher risk of losing muscle mass compared with those without diabetes.42 43 We hypothesise that the relationship between weight loss and an eventual diabetes diagnosis among women may be driven by women who have had long-term, undiagnosed diabetes. This evidence is especially concerning given the higher prevalence of undiagnosed diabetes among women (5.3%) in Indonesia as opposed to men (3.9%) based on data from 2007.44 A recent systematic review found that unintentional weight loss increases mortality risk from all causes, particularly among overweight/obese people and older adult populations.45

Strengths and limitations

This study provides one of the first studies of BMI trajectories and diabetes diagnosis from a middle-income country, and emphasises the importance of longitudinal analysis in exploring the importance of BMI trajectories. Prior work outside of high-income contexts has primarily relied on cross-sectional data or only focused on one sex, providing limited insight into how individual-level BMI change over time influences diabetes onset. This study also relies on biomarker measures of individuals’ weight and height, waist-to-hip ratio and HbA1c level. The analysis of diabetes outcome is obtained from HbA1c level at the last interview, which could bias the effect of age on the probability of diabetes due to underdiagnoses of diabetes at younger ages. However, sensitivity analysis excluding individuals with self-reported physician diagnosed diabetes shows that the associations and direction of covariates remain the same. The effect of initial BMI and BMI change over age are the dominant predictors even in the absence on information of central obesity.

Finally, our study contributes key findings to the literature on BMI change and diabetes from a developing country context that has experienced a rapid increase in overweight/obesity at the population level and has a high rate of underdiagnosed diabetes. Findings from our study confirm evidence from a systematic review of the association between BMI trajectories and diabetes incidence in later life that mostly drew from developed countries.11 Our findings emphasise that simple measurement and monitoring of an individual’s BMI trajectory over adulthood are essential predictors of diabetes onset. Hence, improving public awareness on the effect of BMI trajectory can be useful for prevention and early detection of diabetes risk in the context of a population with substantial unmet need of diabetes screening and a high prevalence of underdiagnosed diabetes. We note that policies on diabetes prevention and screening are mostly targeted at overweight/obese people. However, this study demonstrates the need to identify and target women with unintentional weight loss for diabetes screening. Further, most guidelines recommend screening for diabetes in people aged over 40. Our finding of the different effects of age on the probability of diabetes between men and women highlights a need for screening at earlier ages for women with high BMI trajectories. In fact, a previous study on undiagnosed diabetes in Indonesia estimated that 2.4% of adults at age 28–37 had diabetes but were unaware of their status.44 Similarly, recent studies in the USA and China suggest reducing the age threshold for diabetes screening to age 35 could reduce the proportion of undiagnosed diabetes in the population and act as a low-cost approach to manage implications of diabetes.46 47

Conclusion

We found considerable heterogeneity in the consequences of BMI trajectories by sex for the risk of diabetes in later life. Higher initial BMI and gaining weight over adulthood increase the probability of diabetes in men. For women, our findings suggest the highest risk of diabetes is among those who are initially overweight or obese but experience a rapid loss of BMI over time, a finding that may be explained by long-term, undiagnosed diabetes. This study fills key knowledge gaps by examining the consequences of BMI trajectories for diabetes in a developing country context with rapidly changing diabetes risk factors, a high proportion of underdiagnosed diabetes, and a rapidly ageing population. Our findings highlight that sex-based prevention and screening programmes are necessary to appropriately identify populations at risk and address the rising of diabetes globally.

Overview

Article metrics
Altmetric data not available for this article.
Dimensionsopen-url