This is an open-access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (http://creativecommons.org/licenses/by-nc/4.0/) which permits unrestricted noncommercial use, distribution, and reproduction in any medium, provided the original work is properly cited.
Although obesity is a known risk factor for pancreatic cancer, the role of dynamic weight fluctuations remains unclear. We examined whether body mass index (BMI) variability is associated with pancreatic cancer risk in a large South Korean cohort.
Methods
We analyzed data from 232,322 participants in the Korean National Health Insurance Service database who underwent three health examinations between 2002 and 2007. Participants were followed from 2008 until pancreatic cancer diagnosis, death, or the end of the study period in 2019. BMI variability, assessed using average successive variability, was categorized into tertiles. We used Cox proportional hazards models to calculate adjusted hazard ratios and 95% confidence intervals after adjusting for potential confounders. Sensitivity analyses were performed to verify the robustness of our findings.
Results
In the overall cohort, BMI variability showed no significant association with pancreatic cancer risk. However, a statistically significant interaction by sex was observed (P for interaction=0.029), yielding higher risk estimates among male participants.
Conclusion
Although BMI variability lacked a significant overall association with pancreatic cancer risk, the attenuation of this risk after longer washout periods suggests the potential influence of preclinical weight changes. Nevertheless, the observed effect modification by sex indicates that BMI variability may retain clinical relevance for specific demographic groups, particularly male participants. Therefore, BMI variability should be interpreted not as a direct causal risk factor or intervention target, but as a potential clinical marker that warrants closer monitoring in these specific groups.
Pancreatic cancer ranks among the deadliest cancers, primarily because of its high mortality rate and the absence of reliable early detection methods [1]. Although excess body weight has long been recognized as a major contributor to pancreatic cancer risk [2-4], most epidemiologic studies have conceptualized adiposity as a static exposure, focusing on baseline body mass index (BMI) or net weight gain or loss over time [5-10]. However, the impact of intra-individual body weight variability over time remains understudied. Although weight loss represents an important directional change during the preclinical phase of pancreatic cancer, examining BMI variability captures dynamic weight fluctuations that reflect cumulative metabolic instability beyond net weight change alone.
By potentially triggering repeated metabolic insults and chronic inflammation, BMI variability—independent of net weight gain or loss—may indicate cumulative metabolic stress uncaptured by directional weight trends. These factors underscore the importance of investigating BMI variability not merely as a derivative of obesity, but as an independent factor with distinct pathophysiological implications. In this study, we define BMI variability as intra-individual fluctuations across repeated measurements, regardless of direction.
High BMI variability has been associated with adverse metabolic profiles, including insulin resistance and chronic low-grade inflammation, providing biological plausibility for a link with carcinogenesis [11,12]. However, epidemiological evidence regarding cancer remains limited and inconsistent, partly because of heterogeneous definitions and analytic approaches across studies [13-18].
Among the limited studies evaluating the relationship between weight cycling and pancreatic cancer, the Women’s Health Initiative reported no statistically significant association in postmenopausal female participants [7]. A large US cohort study assessed weight cycling across multiple cancer types but did not provide sex-specific estimates for pancreatic cancer, likely owing to the limited number of cases [19]. The overall association reported in that study was also null.
Because pancreatic cancer lacks established early detection strategies, identifying upstream, potentially modifiable risk patterns—such as those linked to metabolic dysregulation and insulin resistance—is particularly critical. Moreover, the long preclinical phase of pancreatic cancer—during which unintentional weight loss frequently occurs—poses a substantial challenge for causal inference, but it also highlights the potential value of BMI variability as an early clinical signal rather than a causal exposure. Using data from the Korean National Health Insurance Service (NHIS) database, we evaluated whether long-term BMI variability is associated with pancreatic cancer risk beyond static measures of adiposity. We aimed to examine BMI variability as a dynamic weight fluctuation that may reflect underlying metabolic instability or preclinical disease processes, rather than as a direct target for intervention.
Methods
We obtained health screening data from the Korean NHIS database, which provides biennial national health screenings for approximately 97% of the South Korean population, including workers and self-employed individuals aged ≥20 years and dependents aged ≥40 years [20,21]. The study protocol was approved by the Institutional Review Board of Seoul National University Hospital (approval no., X-1701/378-902) and adhered to the ethical standards of the Declaration of Helsinki. Because of the rigorous anonymization of the NHIS database, the requirement for informed consent was waived. Individuals aged ≥40 years who underwent health examinations at three consecutive time points (2002–2003, 2004–2005, and 2006–2007) were initially eligible for this study. Among these individuals, we excluded participants who (1) had missing values for covariates (n=14,694), (2) had a history of cancer before the index date (n=16,820), or (3) died before the index date (n=373). After applying these exclusion criteria, 232,322 participants were included in the final analytic cohort (Figure 1).
Definition of BMI variability
BMI was measured at the three aforementioned time points. To assess BMI variability, we employed the average successive variability (ASV) method, which reflects the average of the absolute differences between consecutive BMI measurements [22]. We categorized participants into tertiles based on their ASV, with the third tertile representing the highest variability and the first tertile representing the lowest. Information regarding intentional versus unintentional weight loss was unavailable in the NHIS database. Consequently, BMI variability reflects observed intraindividual variation and does not distinguish intentional weight control from disease-related or unintentional weight changes.
Definition of outcome
The primary outcome was incident pancreatic cancer. Pancreatic cancer was defined using the International Classification of Diseases, 10th revision (ICD-10) code C25. To ensure accurate case identification, the diagnosis required the ICD-10 code combined with at least two outpatient visits or 1 day of hospitalization. Participants were followed from January 1, 2008, until pancreatic cancer diagnosis, death, or the end of the study period (December 31, 2019).
Statistical analysis
Baseline characteristics were summarized using means and standard deviations for continuous variables, and frequencies and percentages for categorical variables. We calculated adjusted hazard ratios (aHRs) and 95% confidence intervals (CIs) for pancreatic cancer across ASV tertiles using Cox proportional hazards models. We sequentially adjusted for potential confounders across three models. Covariates were selected a priori based on their established or plausible associations with adiposity, weight fluctuations, and pancreatic cancer risk. Model 1 adjusted for age, sex, and household income. Model 2 further adjusted for the Charlson comorbidity index (CCI), baseline BMI, BMI change, systolic blood pressure, fasting serum glucose, total cholesterol, smoking status, alcohol consumption, and physical activity. Model 3 additionally adjusted for metabolic disorders, including hypertension, diabetes, and dyslipidemia.
Stratified analyses were performed in the overall cohort across several covariates: age (<65 or ≥65 years), sex (male or female), smoking status (never- or ever-smokers), weekly alcohol consumption (0 or ≥1 times), weekly physical activity (0 or ≥1 times), hypertension (yes or no), dyslipidemia (yes or no), diabetes (yes or no), CCI (0 or ≥1), baseline BMI (<23.0 or ≥23.0 kg/m²), and direction of BMI change (gain, loss, or stable). To address potential reverse causation, we conducted sensitivity analyses excluding events that occurred within 1 or 2 years of follow-up initiation. Additionally, we categorized participants into ASV deciles to assess the linear relationship between BMI variability and pancreatic cancer risk. We calculated aHRs per one-decile increase in ASV. A two-sided P-value <0.05 was considered statistically significant. All statistical analyses were performed using SAS ver. 9.4 (SAS Institute).
Results
Table 1 summarizes the baseline characteristics of the cohort across BMI variability tertiles. Sex distribution varied across BMI variability tertiles. Participants in the highest BMI variability tertile were older and had a higher baseline BMI, systolic blood pressure, fasting serum glucose, total cholesterol, and CCI. This tertile also included a higher proportion of heavy drinkers and individuals in the lower 50% household income bracket. These findings are consistent with previous reports indicating that high BMI variability is associated with low income and unhealthy behaviors [23].
Table 2 presents the association between BMI variability and pancreatic cancer risk. Overall, the association between BMI variability and pancreatic cancer risk was modest and not statistically significant in the overall cohort. Furthermore, each one-decile increase in BMI variability was not significantly associated with pancreatic cancer risk.
Table 3 presents the results of stratified analyses by key covariates within the overall cohort. Across most subgroups, associations between BMI variability and pancreatic cancer risk were not statistically significant. However, a statistically significant association was observed among male participants (aHR for the highest vs. lowest tertile, 1.26; 95% CI, 1.05–1.51), whereas the corresponding estimates for female participants were not statistically significant. A statistically significant interaction by sex was confirmed (P for interaction=0.029). Among patients with diabetes mellitus, point estimates suggested a higher risk of pancreatic cancer for the highest BMI variability tertile; however, no statistically significant interaction was observed between BMI variability and diabetes (P for interaction >0.05).
Sensitivity analyses using one- and 2-year washout periods are presented in Supplement 1. In the overall cohort, the association between BMI variability and pancreatic cancer risk was attenuated and remained statistically nonsignificant after either washout period.
Among male participants, higher BMI variability remained associated with pancreatic cancer risk after the 1-year washout (aHR for the highest vs. lowest tertile, 1.21; 95% CI, 1.01–1.47; P for trend=0.043), but this association was attenuated and lost statistical significance after the 2-year washout. Among female participants, no statistically significant associations were observed after either washout period. Overall, effect estimates were attenuated with longer washout periods, suggesting the potential influence of reverse causation.
Discussion
Clinically, our findings suggest that pronounced BMI variability should be interpreted not as a direct intervention target, but rather as a potential clinical signal reflecting underlying metabolic instability or early disease-related changes. Given the long preclinical phase of pancreatic cancer and the modest effect size observed, BMI variability is unlikely to represent a direct causal risk factor or a standalone tool for risk stratification. Instead, unexplained or marked BMI variability—particularly in the absence of intentional weight control—warrants closer clinical monitoring rather than direct behavioral intervention.
Previous investigations examining weight variability and pancreatic cancer risk have largely reported null findings in the general population and did not provide sex-stratified estimates, likely owing to a limited number of cases [7,19]. Additionally, most of these studies were conducted in Western populations, predominantly involving White individuals [7,19]. Although obesity or weight gain have been analyzed as contributors to pancreatic cancer risk using sex-stratified analyses [4], none have specifically evaluated BMI variability in this context. Because pancreatic cancer risk and metabolic profiles vary across demographic and clinical subgroups, we evaluated sex as a potential effect modifier within the overall cohort to explore heterogeneity in the association between BMI variability and pancreatic cancer risk.
We emphasize that sex was evaluated strictly as an exploratory effect modifier rather than a defining feature of the study design. Differences in baseline risk profiles—such as higher rates of smoking, alcohol consumption, and incident diabetes and pancreatic cancer among male participants compared with female participants [24,25]—may contextualize the variation in effect estimates by sex; however, these differences do not establish a causal explanation. The nonsignificant findings among female participants are consistent with previous studies reporting null associations [7,19]. Therefore, these female-specific results must be viewed within the context of subgroup analyses with limited case numbers, and all sex-specific findings in this cohort should be considered exploratory rather than definitive.
In exploratory subgroup analyses, a similar pattern emerged among patients with diabetes mellitus, a known high-risk group for pancreatic cancer. However, because the interaction by diabetes was not statistically significant, this finding should be interpreted strictly as a descriptive observation rather than evidence of true effect modification.
Reverse causation remains a critical consideration, as unintentional weight loss during the preclinical phase of pancreatic cancer inherently increases BMI variability. Although washout analyses attenuated the risk estimates, a positive trend persisted, suggesting that this association cannot be fully explained by preclinical disease alone. After applying a 1-year washout period, the association remained statistically significant among male participants, whereas further attenuation occurred after a 2-year washout and among female participants. Nevertheless, reverse causation cannot be definitively excluded, and future studies with extended follow-up periods are required to clarify the clinical significance of BMI variability.
Therefore, BMI variability may function as a clinical marker or proxy for underlying metabolic disturbances, rather than as an independent driver of pancreatic carcinogenesis. Understanding the impact of BMI variability on pancreatic cancer risk is essential given the increasing burdens of both obesity and pancreatic cancer [25,26].
One possible pathway linking BMI variability to pancreatic cancer risk involves its relationship with central and visceral adiposity. Repeated weight gain and loss may preferentially promote abdominal fat accumulation, even without substantial changes in overall BMI. Central obesity, which is more strongly associated with pancreatic carcinogenesis than general obesity [2,27,28], has been linked to higher pancreatic cancer risk [3] and is closely associated with BMI variability. Visceral fat, a major component of central obesity, is metabolically active and known to promote metabolic dysfunction and chronic inflammation—mechanisms heavily implicated in pancreatic carcinogenesis [29].
Higher BMI variability has been associated with metabolic disturbances such as inflammation, impaired energy balance, and insulin resistance. However, these mechanistic links remain exploratory and do not imply that modifying BMI variability itself would necessarily translate into reduced pancreatic cancer risk.
Our study has several strengths. First, we utilized a large-scale national health insurance database to examine pancreatic cancer risk in a population-based setting. Second, we adjusted for multiple risk factors to enhance analytical rigor. Third, we used the ASV method, a validated measure of BMI variability that sensitively captures intra-individual changes in BMI over time. Finally, unlike prior studies that focused on unidirectional weight changes, our study examined BMI variability as a direction-independent weight fluctuation and found suggestive associations with pancreatic cancer risk that warrant cautious interpretation. Furthermore, we conducted sensitivity analyses excluding cases diagnosed within the first 2 years of follow-up to assess the potential impact of reverse causation.
This study also has several limitations. First, intentional and unintentional weight changes could not be distinguished using the available database, limiting causal interpretation and leaving the possibility of reverse causation—particularly given the long preclinical phase of pancreatic cancer, despite the application of washout periods. Second, BMI change was simplified into an increase versus a decrease, which may not fully capture complex weight trajectories; more detailed classifications were not feasible owing to limited sample sizes within subgroups. Third, because this was a retrospective observational study, causal inference cannot be established, and residual confounding, selection bias, and misclassification remain possible. Fourth, although sex-specific differences were observed, the underlying mechanisms could not be elucidated. Furthermore, the absence of detailed body composition measures (e.g., waist circumference or body fat percentage) and the relatively small number of pancreatic cancer cases—particularly among female participants—may have limited our statistical power. Finally, pancreatic cancer was identified using ICD-10 diagnostic codes and healthcare utilization records. Linkage to special reimbursement (catastrophic illness) codes was not feasible owing to data limitations, which may have resulted in some degree of outcome misclassification. Future studies with more refined outcome ascertainment are warranted.
In this large cohort of South Korean adults, high BMI variability showed no statistically significant association with pancreatic cancer risk in the overall cohort. The attenuation of these associations after longer washout periods suggests that preclinical weight changes may have contributed to the observed findings. Nevertheless, effect estimates differed across subgroups, indicating that BMI variability may retain clinical relevance for specific demographic groups. Therefore, BMI variability should be interpreted not as a direct intervention target, but rather as a potential clinical marker warranting further investigation. Future studies incorporating extended follow-up periods and distinguishing intentional from unintentional weight changes are required to clarify the exact nature of this relationship.
Notes
Conflict of interest
No potential conflict of interest relevant to this article was reported.
Funding
This work was supported by the SNUH Research Fund (Grant No. 0320250220).
Data availability
The data were obtained from the NHIS of Korea for approved research use, and the authors do not have the authority to share them.
Values are presented as number (%), mean±standard deviation, or mean (range). ASV is defined as the average of the absolute differences between the 1st and 2nd BMI measurements and between the 2nd and 3rd BMI measurements.
BMI, body mass index; ASV, average successive variability.
Table 2.
Association between BMI variability and pancreatic cancer risk
Values are presented as adjusted hazard ratio (95% confidence interval). ASV is defined as the average of the absolute differences between the 1st and 2nd BMI measurements and between the 2nd and 3rd BMI measurements.
BMI, body mass index; ASV, average successive variability; aHR, adjusted hazard ratio; CI, confidence interval.
a)Model 1 adjusted for age, sex, and household income.
b)Model 2 further adjusted for Charlson comorbidity index, BMI change, BMI, systolic blood pressure, fasting serum glucose, and total cholesterol, smoking status, alcohol consumption, physical activity on the basis of Model 1.
c)Model 3 further adjusted for metabolic disorders, including hypertension, diabetes, and dyslipidemia on the basis of Model 2.
Table 3.
Stratified analyses of the association between BMI variability and pancreatic cancer risk
Characteristic
Total
Event
BMI variability (ASV group)
P for trend
P for interaction
1st tertile
2nd tertile
3rd tertile
Sex
0.029
Male
136,365
696
1.00 (Reference)
1.03 (0.85–1.24)
1.26 (1.05–1.51)
0.012
Female
95,957
414
1.00 (Reference)
1.06 (0.83–1.35)
0.93 (0.73–1.18)
0.484
Age (y)
0.345
<65
188,699
679
1.00 (Reference)
1.10 (0.91–1.33)
1.17 (0.97–1.41)
0.101
≥65
43,623
431
1.00 (Reference)
0.97 (0.76–1.23)
1.09 (0.87–1.38)
0.412
Smoking status
0.219
Never smoker
151,791
698
1.00 (Reference)
1.11 (0.92–1.33)
1.06 (0.88–1.28)
0.572
Ever smoker
80,531
412
1.00 (Reference)
0.94 (0.74–1.21)
1.24 (0.98–1.57)
0.063
Alcohol consumption (times/wk)
0.234
0
164,322
773
1.00 (Reference)
1.00 (0.84–1.20)
1.06 (0.89–1.26)
0.521
≥1
68,000
337
1.00 (Reference)
1.14 (0.87–1.50)
1.29 (0.98–1.68)
0.066
Physical activity (times/wk)
0.249
0
124,176
607
1.00 (Reference)
1.05 (0.85–1.29)
1.21 (0.99–1.47)
0.055
≥1
108,146
503
1.00 (Reference)
1.05 (0.85–1.30)
1.02 (0.82–1.27)
0.831
Hypertension
0.787
No
162,195
654
1.00 (Reference)
1.09 (0.90–1.31)
1.13 (0.94–1.37)
0.201
Yes
70,127
456
1.00 (Reference)
0.99 (0.78–1.25)
1.10 (0.88–1.38)
0.383
Dyslipidemia
0.387
No
204,609
961
1.00 (Reference)
1.05 (0.89–1.23)
1.10 (0.94–1.29)
0.236
Yes
27,713
149
1.00 (Reference)
1.02 (0.67–1.56)
1.26 (0.85–1.88)
0.237
Diabetes mellitus
0.090
No
213,451
942
1.00 (Reference)
1.05 (0.89–1.23)
1.06 (0.90–1.24)
0.484
Yes
18,871
168
1.00 (Reference)
1.03 (0.67–1.57)
1.51 (1.03–2.22)
0.022
Charlson comorbidity index
0.151
0
52,636
194
1.00 (Reference)
1.03 (0.72–1.47)
1.41 (1.00–1.99)
0.051
≥1
179,686
916
1.00 (Reference)
1.05 (0.89–1.23)
1.08 (0.92–1.27)
0.355
Initial BMI (kg/m2)
0.826
<23.0
87,645
418
1.00 (Reference)
1.13 (0.89–1.42)
1.09 (0.86–1.39)
0.476
≥23.0
144,677
692
1.00 (Reference)
1.00 (0.82–1.21)
1.14 (0.95–1.37)
0.140
Direction of BMI change
0.903
Gain
111,212
521
1.00 (Reference)
1.08 (0.87–1.35)
1.10 (0.88–1.36)
0.409
Loss
108,596
534
1.00 (Reference)
1.01 (0.82–1.26)
1.15 (0.93–1.42)
0.182
Stable
12,514
55
1.00 (Reference)
0.99 (0.50–1.94)
1.15 (0.60–2.18)
0.712
Values are presented as adjusted hazard ratio (95% confidence interval). ASV is defined as the average of the absolute differences between the 1st and 2nd BMI measurements and between the 2nd and 3rd BMI measurements. Model adjusted for age, sex, household income, Charlson comorbidity index, BMI change, BMI, systolic blood pressure, fasting serum glucose, total cholesterol, smoking status, alcohol consumption, physical activity, hypertension, diabetes, and dyslipidemia.
BMI, body mass index; ASV, average successive variability.
References
1. US Preventive Services Task Force; Owens DK, Davidson KW, Krist AH, Barry MJ, Cabana M, et al. Screening for pancreatic cancer: US Preventive Services Task Force reaffirmation recommendation statement. JAMA 2019;322:438-44.
2. Aune D, Greenwood DC, Chan DS, Vieira R, Vieira AR, Navarro Rosenblatt DA, et al. Body mass index, abdominal fatness and pancreatic cancer risk: a systematic review and non-linear dose–response meta-analysis of prospective studies. Ann Oncol 2012;23:843-52.
3. Arslan AA, Helzlsouer KJ, Kooperberg C, Shu XO, Steplowski E, Bueno-de-Mesquita HB, et al. Anthropometric measures, body mass index, and pancreatic cancer: a pooled analysis from the pancreatic cancer Cohort Consortium (PanScan). Arch Intern Med 2010;170:791-802.
4. Patel AV, Rodriguez C, Bernstein L, Chao A, Thun MJ, Calle EE. Obesity, recreational physical activity, and risk of pancreatic cancer in a large US cohort. Cancer Epidemiol Biomarkers Prev 2005;14:459-66.
5. Stolzenberg-Solomon RZ, Schairer C, Moore S, Hollenbeck A, Silverman DT. Lifetime adiposity and risk of pancreatic cancer in the NIHAARP Diet and Health Study cohort. Am J Clin Nutr 2013;98:1057-65.
6. da Silva M, Weiderpass E, Licaj I, Lissner L, Rylander C. Excess body weight, weight gain and obesity-related cancer risk in women in Norway: the Norwegian Women and Cancer study. Br J Cancer 2018;119:646-56.
7. Schwalb ME, Smith-Warner SA, Hou J, Rohan TE, Snetselaar L, Luo J, et al. Sustained weight loss, weight cycling, and weight gain during adulthood and pancreatic cancer incidence in the Women’s Health Initiative. Am J Epidemiol 2022;191:1009-20.
10. Jiao L, Berrington de Gonzalez A, Hartge P, Pfeiffer RM, Park Y, Freedman DM, et al. Body mass index, effect modifiers, and risk of pancreatic cancer: a pooled study of seven prospective cohorts. Cancer Causes Control 2010;21:1305-14.
13. Lindblad P, Wolk A, Bergstrom R, Persson I, Adami HO. The role of obesity and weight fluctuations in renal cell cancer. Cancer Epidemiol Biomarkers Prev 1994;3:631-9.
14. French SA, Folsom AR, Jeffery RW, Zheng W, Mink PJ, Baxter JE. Weight variability and incident disease in older women: the Iowa Women’s Health Study. Int J Obes Relat Metab Disord 1997;21:217-23.
15. Cho YJ, Kawk JS, Yoon HJ, Park M. Body weight variability and cancer incidence in men aged 40 years and older: Korean National Insurance Service Cohort. Sci Rep 2021;11:12122.
17. Eng SM, Gammon MD, Terry MB, Kushi LH, Teitelbaum SL, Britton JA, et al. Body size changes in relation to postmenopausal breast cancer. Am J Epidemiol 2005;162:229-37.
18. De Roos AJ, Ulrich CM, Ray RM, Mossavar-Rahmani Y, Rosenberg CA, Caan BJ, et al. Intentional weight loss and risk of lymphohematopoietic cancers. Cancer Causes Control 2010;21:223-36.
19. Stevens VL, Jacobs EJ, Patel AV, Sun J, McCullough ML, Campbell PT, et al. Weight cycling and cancer incidence in a large prospective US cohort. Am J Epidemiol 2015;182:394-404.
20. Seong SC, Kim YY, Park SK, Khang YH, Kim HC, Park JH, et al. Cohort profile: the National Health Insurance Service–National Health Screening Cohort. BMJ Open 2017;7:e016640.
21. Cheol Seong S, Kim YY, Khang YH, Heon Park J, Kang HJ, Lee H, et al. Data resource profile: the National Health Information Database of South Korea. Int J Epidemiol 2017;46:799-800.
26. The Lancet Gastroenterology Hepatology. Cause for concern: the rising incidence of early-onset pancreatic cancer. Lancet Gastroenterol Hepatol 2023;8:287.
27. Cereda E, Malavazos AE, Caccialanza R, Rondanelli M, Fatati G, Barichella M. Weight cycling is associated with abdominal fat accumulation. Clin Nutr 2011;30:718-23.
28. Barberio AM, Alareeki A, Viner B, Pader J, Vena JE, Arora P, et al. Central body fatness is a stronger predictor of cancer risk. Nat Commun 2019;10:383.
Body mass index variability and pancreatic cancer risk in South Korean adults: a nationwide cohort study
Figure. 1. Flowchart of the study cohort.
Graphical abstract
Figure. 1.
Graphical abstract
Body mass index variability and pancreatic cancer risk in South Korean adults: a nationwide cohort study
Characteristic
Total population
BMI variability (ASV)
1st tertile
2nd tertile
3rd tertile
No. of participants
232,322
77,111
77,844
77,367
No. of events
1,110
334
364
412
Sex
Male
136,365 (58.7)
48,253 (62.6)
46,213 (59.4)
41,899 (54.2)
Female
95,957 (41.3)
28,858 (37.4)
31,631 (40.6)
35,468 (45.9)
Age (y)
56.09±8.79
55.36±8.39
55.92±8.66
57.00±9.23
Weight variability (ASV) (kg/m2)
0.88 (0.00–36.65)
0.33 (0.00–0.53)
0.73 (0.54–0.96)
1.57 (0.96–36.65)
Initial BMI (kg/m2)
23.97±2.89
23.67±2.67
23.86±2.75
24.38±3.18
Change in BMI (kg/m2)
0.01±1.47
0.01±0.58
0.04±1.11
–0.03±2.22
Gain
111,212 (47.9)
35,719 (46.3)
38,442 (49.4)
37,051 (47.9)
Loss
108,596 (46.7)
34,518 (44.8)
36,441 (46.8)
37,637 (48.7)
Stable
12,514 (5.4)
6,874 (9.0)
2,961 (3.8)
2,679 (3.5)
Systolic blood pressure (mm Hg)
126.37±17.46
125.66±17.10
126.14±17.35
127.30±17.88
Fasting serum glucose (mg/dL)
96.66±31.12
95.77±28.76
96.37±30.72
97.83±33.65
Total cholesterol (mg/dL)
200.16±37.58
199.41±37.11
199.96±37.43
201.12±38.17
Household income
1st quartile (lowest)
29,609 (12.7)
8,786 (11.4)
9,703 (12.5)
11,120 (14.4)
2nd quartile
46,058 (19.8)
13,967 (18.1)
15,287 (19.6)
16,804 (21.7)
3rd quartile
66,642 (28.7)
21,652 (28.1)
22,374 (28.7)
22,616 (29.2)
4th quartile (highest)
90,013 (38.7)
32,706 (42.4)
30,480 (39.2)
26,827 (34.7)
Smoking status
Never smoker
151,791 (65.3)
49,253 (63.9)
50,726 (65.2)
51,812 (67.0)
Past smoker
23,184 (10.0)
8,344 (10.8)
7,859 (10.1)
6,981 (9.0)
Current smoker
57,347 (24.7)
19,514 (25.3)
19,259 (24.7)
18,574 (24.0)
Alcohol consumption frequency (times/wk)
None
164,322 (70.7)
53,241 (69.0)
54,814 (70.4)
56,267 (72.7)
1–2
42,732 (18.4)
15,317 (19.9)
14,506 (18.6)
12,909 (16.7)
3–4
16,365 (7.0)
5,693 (7.4)
5,539 (7.1)
5,133 (6.6)
≥5
8,903 (3.8)
2,860 (3.7)
2,985 (3.8)
3,058 (4.0)
Physical activity frequency (times/wk)
None
124,176 (53.5)
39,122 (50.7)
41,313 (53.1)
43,741 (56.5)
1–2
62,016 (26.7)
22,035 (28.6)
20,958 (26.9)
19,023 (24.6)
3–4
24,575 (10.6)
8,882 (11.5)
8,328 (10.7)
7,365 (9.5)
≥5
21,555 (9.3)
7,072 (9.2)
7,245 (9.3)
7,238 (9.4)
Charlson comorbidity index
0
52,636 (22.7)
18,806 (24.4)
17,911 (23.0)
15,919 (20.6)
1
61,852 (26.6)
21,490 (27.9)
21,009 (27.0)
19,353 (25.0)
≥2
117,834 (50.7)
36,815 (47.7)
38,924 (50.0)
42,095 (54.4)
BMI variability (ASV group)
Participant
Event
Person-years
Model 1a)
Model 2b)
Model 3c)
1st tertile
77,111
334
899,478
1.00 (Reference)
1.00 (Reference)
1.00 (Reference)
2nd tertile
77,844
364
904,487
1.05 (0.91–1.22)
1.05 (0.90–1.21)
1.04 (0.90–1.21)
3rd tertile
77,367
412
888,100
1.15 (0.99–1.33)
1.13 (0.97–1.30)
1.12 (0.97–1.30)
P for trend
-
-
-
0.059
0.112
0.121
aHR (95% CI) per 1-decile ASV increase
-
-
-
1.01 (0.99–1.03)
1.01 (0.99–1.03)
1.01 (0.99–1.03)
Characteristic
Total
Event
BMI variability (ASV group)
P for trend
P for interaction
1st tertile
2nd tertile
3rd tertile
Sex
0.029
Male
136,365
696
1.00 (Reference)
1.03 (0.85–1.24)
1.26 (1.05–1.51)
0.012
Female
95,957
414
1.00 (Reference)
1.06 (0.83–1.35)
0.93 (0.73–1.18)
0.484
Age (y)
0.345
<65
188,699
679
1.00 (Reference)
1.10 (0.91–1.33)
1.17 (0.97–1.41)
0.101
≥65
43,623
431
1.00 (Reference)
0.97 (0.76–1.23)
1.09 (0.87–1.38)
0.412
Smoking status
0.219
Never smoker
151,791
698
1.00 (Reference)
1.11 (0.92–1.33)
1.06 (0.88–1.28)
0.572
Ever smoker
80,531
412
1.00 (Reference)
0.94 (0.74–1.21)
1.24 (0.98–1.57)
0.063
Alcohol consumption (times/wk)
0.234
0
164,322
773
1.00 (Reference)
1.00 (0.84–1.20)
1.06 (0.89–1.26)
0.521
≥1
68,000
337
1.00 (Reference)
1.14 (0.87–1.50)
1.29 (0.98–1.68)
0.066
Physical activity (times/wk)
0.249
0
124,176
607
1.00 (Reference)
1.05 (0.85–1.29)
1.21 (0.99–1.47)
0.055
≥1
108,146
503
1.00 (Reference)
1.05 (0.85–1.30)
1.02 (0.82–1.27)
0.831
Hypertension
0.787
No
162,195
654
1.00 (Reference)
1.09 (0.90–1.31)
1.13 (0.94–1.37)
0.201
Yes
70,127
456
1.00 (Reference)
0.99 (0.78–1.25)
1.10 (0.88–1.38)
0.383
Dyslipidemia
0.387
No
204,609
961
1.00 (Reference)
1.05 (0.89–1.23)
1.10 (0.94–1.29)
0.236
Yes
27,713
149
1.00 (Reference)
1.02 (0.67–1.56)
1.26 (0.85–1.88)
0.237
Diabetes mellitus
0.090
No
213,451
942
1.00 (Reference)
1.05 (0.89–1.23)
1.06 (0.90–1.24)
0.484
Yes
18,871
168
1.00 (Reference)
1.03 (0.67–1.57)
1.51 (1.03–2.22)
0.022
Charlson comorbidity index
0.151
0
52,636
194
1.00 (Reference)
1.03 (0.72–1.47)
1.41 (1.00–1.99)
0.051
≥1
179,686
916
1.00 (Reference)
1.05 (0.89–1.23)
1.08 (0.92–1.27)
0.355
Initial BMI (kg/m2)
0.826
<23.0
87,645
418
1.00 (Reference)
1.13 (0.89–1.42)
1.09 (0.86–1.39)
0.476
≥23.0
144,677
692
1.00 (Reference)
1.00 (0.82–1.21)
1.14 (0.95–1.37)
0.140
Direction of BMI change
0.903
Gain
111,212
521
1.00 (Reference)
1.08 (0.87–1.35)
1.10 (0.88–1.36)
0.409
Loss
108,596
534
1.00 (Reference)
1.01 (0.82–1.26)
1.15 (0.93–1.42)
0.182
Stable
12,514
55
1.00 (Reference)
0.99 (0.50–1.94)
1.15 (0.60–2.18)
0.712
Table 1. Descriptive characteristics of the study cohort
Values are presented as number (%), mean±standard deviation, or mean (range). ASV is defined as the average of the absolute differences between the 1st and 2nd BMI measurements and between the 2nd and 3rd BMI measurements.
BMI, body mass index; ASV, average successive variability.
Table 2. Association between BMI variability and pancreatic cancer risk
Values are presented as adjusted hazard ratio (95% confidence interval). ASV is defined as the average of the absolute differences between the 1st and 2nd BMI measurements and between the 2nd and 3rd BMI measurements.
BMI, body mass index; ASV, average successive variability; aHR, adjusted hazard ratio; CI, confidence interval.
Model 1 adjusted for age, sex, and household income.
Model 2 further adjusted for Charlson comorbidity index, BMI change, BMI, systolic blood pressure, fasting serum glucose, and total cholesterol, smoking status, alcohol consumption, physical activity on the basis of Model 1.
Model 3 further adjusted for metabolic disorders, including hypertension, diabetes, and dyslipidemia on the basis of Model 2.
Table 3. Stratified analyses of the association between BMI variability and pancreatic cancer risk
Values are presented as adjusted hazard ratio (95% confidence interval). ASV is defined as the average of the absolute differences between the 1st and 2nd BMI measurements and between the 2nd and 3rd BMI measurements. Model adjusted for age, sex, household income, Charlson comorbidity index, BMI change, BMI, systolic blood pressure, fasting serum glucose, total cholesterol, smoking status, alcohol consumption, physical activity, hypertension, diabetes, and dyslipidemia.
BMI, body mass index; ASV, average successive variability.
, Ju Hyun Kang2
