Body mass index trajectories among people with obesity and association with mortality: Evidence from a large Israeli database

Abstract Objective Previous studies using longitudinal weight data to characterize obesity are based on populations of limited size and mostly include individuals of all body mass index (BMI) levels, without focusing on weight changes among people with obesity. This study aimed to identify BMI trajectories over 5 years in a large population with obesity, and to determine the trajectories' association with mortality. Methods For inclusion, individuals aged 30–74 years at index date (1 January 2013) with continuous membership in Clalit Health Services from 2008 to 2012 were required to have ≥1 BMI measurement per year in ≥3 calendar years during this period, of which at least one was ≥30 kg/m2. Latent class analysis was used to generate BMI trajectories over 5 years (2008–2012). Cox proportional hazards models were used to assess the association between BMI trajectories and all‐cause mortality during follow‐up (2013–2017). Results In total, 367,141 individuals met all inclusion criteria. Mean age was 57.2 years; 41% were men. The optimal model was a quadratic model with four classes of BMI clusters. Most individuals (90.0%) had stable high BMI over time. Individuals in this cluster had significantly lower mortality than individuals in the other trajectory clusters (p < 0.01), including clusters of people with dynamic weight trajectories. Conclusions The results of the current study show that people with stable high weight had the lowest mortality of all four BMI trajectories identified. These findings help to expand the scientific understanding of the impact that weight trajectories have on health outcomes, while demonstrating the challenges of discerning the cumulative effects of obesity and weight change, and suggest that dynamic historical measures of BMI should be considered when assessing patients' future risk of obesity‐related morbidity and mortality, and when choosing a treatment strategy.


| INTRODUCTION
The association between individuals' increasing high body mass index (BMI) and higher rates of clinical complications is well established. 1 Various studies have shown an increase in mortality with increasing BMI [2][3][4] ; furthermore, individuals with overweight or obesity are at greater risk of developing diseases such as type 2 diabetes (T2D) 5 or cardiovascular (CV) disease 5-7 than individuals with healthy weight.
Consequently, a new international classification of disease (ICD) for obesity was recently proposed, based on a combination of pathophysiology, BMI, CV complications remediable by weight loss, and the severity of complications. 8 Obesity-related complications are also linked to increased mortality, 9 and high direct and indirect healthcare costs. 10,11 According to international guidelines, BMI level and the presence of comorbidities are key criteria in treatment decisions for the management of obesity. 12 However, much of the available evidence on obesityrelated complications is based on BMI measurements calculated at one time point, whereas the relationship between weight change over time and obesity-related complications is less well characterized. Outside interventional clinical trials, studies examining how gaining, maintaining or losing weight differentially affects the risks for obesity-related outcomes have yielded contradictory results.
Several studies have shown an association between stable high or increasing weight trajectories and increased CV risk 13 and mortality. 14,15 However, others have reported that fluctuating weight is linked to comparatively poorer outcomes in patients with T2D, 16 and carries an overall greater CV risk 17,18 and higher mortality 18 than stable weight.
In clinical trials and observational studies, weight loss is typically studied in the context of a specific medical 19,20 or surgical 21,22 intervention for the management of obesity over a certain time frame. Other approaches are required to provide information on weight change in a general population with obesity. However, accurate identification of weight trajectories based on real-world data can be limited by both the breadth and the depth of patient data available for longitudinal analysis. Several previous studies have assessed longitudinal weight trajectories in adults and their association with selected health outcomes, but these have also been subject to some of the common limitations of observational studies. For example, many included relatively small populations, 14,[23][24][25][26][27] selected individuals within a narrow age range, 14,25 or defined BMI trajectories for the general population rather than specifically for individuals with obesity. 14,[23][24][25][26][27][28] In the absence of extensive evidence, assessment of longitudinal BMI data in real-world populations with obesity is highly valuable, both contributing to a more accurate clinical description of obesity phenotypes and strengthening the understanding of the impact of weight changes over time. The main objective of this study was to identify BMI trajectories over a 5-year period in a large population with obesity, using a comprehensive Israeli electronic health record database, and to assess the association of BMI trajectories with allcause mortality.

| Data source
This study used historical data from Clalit Health Services, 29

| Study design and population
This was a retrospective cohort study, with the following dates and study periods ( Figure 1):  antidepressants) and weight-loss interventions (medication, surgical intervention, and visit to a dietician) during the 5 years pre-index date were also recorded.

| Outcomes
The primary outcome was all-cause mortality (yes/no only; cause not defined) during the 5 years post-index date. Current and complete information regarding mortality events was obtained from Israel's Ministry of the Interior, which includes the entire Israeli population.

Secondary outcomes included incidence of new diagnosis of T2D
(defined using a Clalit Research Institute-reported algorithm 31 ), major adverse cardiac events (MACE; defined as incidence of myocardial infarction, unstable angina pectoris, any percutaneous transluminal coronary angioplasty or any coronary artery bypass graft) and chronic kidney disease.

| Statistical analysis
The main characteristics of the total study population were described using proportions for categorical variables and means with standard deviation (SD) for continuous variables. To assess the unadjusted and adjusted associations between BMI (last measure and identified trajectories) and all-cause mortality, a fixed Cox regression, with baseline (last) BMI and the BMI trajectory cluster as the main exposure, was used, with adjustment for potential confounders in the baseline variable collection period, including age; sex; immigration status; ethnicity; socioeconomic status; place of residence; marital status; and comorbidities at index date (see Figure 4 for complete list). Termination of follow-up was defined as death or end of followup period (31 December 2017). Individuals with known active cancer during 2008-2012 were excluded from this analysis. The assumption of proportional hazards was tested and a p value <0.05 with twosided test was used as the statistically significant threshold.

| Ethical approval and use of data
This study using secondary data was approved by Clalit Health Services' institutional review board, in accordance with the Declaration of Helsinki. Data were used in accordance with the terms agreed to upon their receipt.  Table 1. The mean age was 57.2 years (SD 10.8), the majority were women (59.0%) and mean Charlson comorbidity score was 1.5 (SD 1.7). Compared with those excluded from the study owing to missing BMI measurements, the study population was significantly older (57.2 vs. 44.2 years; p < 0.001) and had more comorbidities (Charlson score: 1.53 vs. 0.47; p < 0.001), and a greater proportion were women (59.0% vs. 52.1%; p < 0.001; Table 2).

| BMI measurements
A total of 2,846,323 valid BMI measurements were recorded for the study population over the 5-years trajectory categorization period, with a mean number of measurements per person of 7.75 (SD 5.76) for the entire 5 years (Table 3). The mean first BMI measurement was similar to the mean of all other BMI measurements taken (∼33 kg/m 2 across all other BMI-related variables), demonstrating its robustness for use in this analysis.

| BMI trajectories
A total of 10 different LCA models (five linear and five quadratic models) were used to examine the performance of 2-6 clusters for each type. The optimal model, with AIC 7,695,024.91 and BIC 7,695,232.11, was a quadratic model with four distinct clusters of BMI trajectories ( Table 4).
Most individuals with obesity at baseline displayed stable high BMI over time (330,558; 90.0% of the study population). A further 28,907 individuals (7.9%) had very high, slightly increasing BMI over the study period. Very few individuals had dynamic increasingdecreasing BMI (4889; 1.3%) and an even smaller proportion had dynamic decreasing-increasing BMI (2787; 0.8%) over time ( Figure 3).

| Baseline characteristics across BMI trajectories
Baseline characteristics across BMI trajectories are shown in Table 1  -153

| BMI trajectories and all-cause mortality during the follow-up period
The highest incidence of all-cause mortality during the follow-up period was observed among individuals in the very high, slightly increasing BMI trajectory cluster (7.2%), followed by 6.3% in the dynamic increasing-decreasing BMI trajectory cluster, 5.2% in the dynamic decreasing-increasing BMI trajectory cluster and 4.6% in the stable high BMI trajectory cluster (

| Secondary outcomes
During the follow-up period there was a higher incidence of T2D  Table 6.

| DISCUSSION
This retrospective cohort study utilized population-based electronic health record data from 367,141 adults with obesity. Four BMI trajectories over a 5-year period were generated using LCA, and the stable high weight trajectory, consisting of 90% of individuals, was found to be associated with lower mortality compared with very high, slightly increasing or dynamic BMI trajectories. The finding that most individuals with obesity maintained a stable high BMI supports previous studies conducted in the general population, which have described weight loss in fewer than 5% of individuals, whereas 70%-90% maintained a stable weight. 25,26 These findings are also in accord with studies indicating an association between weight stability and lower mortality. 25,36,37 This could be because fewer individuals with stable high BMI experienced severe obesity (BMI >40 kg/m 2 ), compared with those in other trajectory clusters. Alternatively, it may reflect differences between trajectories in terms of total cumulative time with obesity and the prevalence of comorbidities. The higher risk of all-cause mortality among individuals with dynamic BMI may also result from BMI changes triggered by an underlying disease, which is suggested by the small proportion of patients receiving therapeutic weight-loss treatment in this study, because unintentional weight loss often indicates that a patient may have a serious, hitherto undiagnosed, condition.
Evidence on how weight loss affects morbidity and mortality is mixed. Several studies have reported that weight loss and accompanying weight fluctuations can lead to increased morbidity 16-18 ; however, in a UK epidemiological study, individuals who had lost and regained weight had a lower CV risk than those with stable obesity, overweight or normal weight, suggesting that weight loss, even if not sustained, could result in long-term CV benefit. 38 Similarly, intentional weight loss in clinical trials is often associated with decreased mortality, 19 but a link between weight loss and increased mortality has been previously reported in an observational study. 39  Abbreviations: BMI, body mass index; CI, confidence interval; CKD, chronic kidney disease; IHD, ischemic heart disease; MACE; major adverse cardiac events; RTT, renal replacement therapy; T2D, type 2 diabetes.