Body mass index and body shape before treatment and nasopharyngeal carcinoma prognosis: A population‐based patient cohort study in southern China

A concern of reverse causation exists about the association between nasopharyngeal carcinoma (NPC) prognosis and body mass index (BMI) at diagnosis, while the prognostic impact of BMI measured years before diagnosis is unknown. Therefore, we investigated associations of prediagnosis and pretreatment BMI and body shape on NPC mortality. From a population‐based patient cohort in southern China between 2010 and 2013, we included 2526 incident NPC cases with prospective follow‐up through 2018. We assessed the associations of BMI and body shape at age 20 years, 10 years before diagnosis, and at diagnosis with NPC mortality, combining strategies of stratification and statistical adjustment to minimize reverse causation. We observed 25% lower all‐cause mortality (hazard ratio [HR] 0.75, 95% confidence interval [CI]: 0.64‐0.89) and 25% lower NPC‐specific mortality (HR 0.75, 95% CI: 0.61‐0.91) among overweight vs normal‐weight NPC cases at diagnosis. Lean body shapes 1 and 2 at diagnosis were associated with 68% and 23% higher all‐cause mortality, respectively, compared to normal body shape 3. No effect modification by cancer stage was detected for associations with all‐cause or NPC‐specific mortality. Associations with BMI and body shape 10 years before diagnosis were similar but attenuated, while body size and shape at age 20 were not associated with mortality. Being overweight at diagnosis decreased mortality, and thinner body shape increased mortality, compared to normal weight/body shape. These associations may be due to poorer nutrition and treatment intolerance, resulting in treatment discontinuation and worse survival outcomes.

weight NPC cases at diagnosis. Lean body shapes 1 and 2 at diagnosis were associated with 68% and 23% higher all-cause mortality, respectively, compared to normal body shape 3. No effect modification by cancer stage was detected for associations with allcause or NPC-specific mortality. Associations with BMI and body shape 10 years before diagnosis were similar but attenuated, while body size and shape at age 20 were not associated with mortality. Being overweight at diagnosis decreased mortality, and thinner body shape increased mortality, compared to normal weight/body shape. These associations may be due to poorer nutrition and treatment intolerance, resulting in treatment discontinuation and worse survival outcomes.

K E Y W O R D S
body mass index, body shape, nasopharyngeal carcinoma, prognosis, residual confounding, reverse causation What's new?
Body mass index (BMI) is suspected of influencing not only nasopharyngeal carcinoma (NPC) risk but also mortality after NPC diagnosis. Nonetheless, little is known about the impact of BMI at age 20 years and 10 years before NPC diagnosis and mortality. Here, in a population-based patient cohort in southern China, the authors investigated associations between BMI and body shape at 20 years, 10 years before diagnosis and before treatment and NPC mortality. NPC mortality was decreased among patients who were overweight at diagnosis, whereas increased mortality was observed among thinner patients. Prediagnosis BMI and body shape had similar but weaker associations. Lean and underweight NPC patients may benefit from heightened clinical attention to undernutrition and poorer treatment tolerance.

Nasopharyngeal carcinoma (NPC) is endemic in Southeast and East
Asia, North Africa, and the Middle East. 1 The highest incidence is observed in southern China, recently reaching 25.0/100 000 personyears among males and 7.7/100 000 person-years among females. 2 High body mass index (BMI), a measurement of overall body fatness, might increase NPC risk. 3,4 Conversely, risk of mortality after NPC diagnosis may be lower in patients with higher BMI compared to underweight and normal weight patients. [5][6][7] Concerns about reverse causality and residual confounding, however, cast doubt on the observed inverse prognostic association 8 because most previous studies 6,[9][10][11] used BMI measured a substantial time after diagnosis. Low BMI might result from cancer progression and cachexia if patients are diagnosed at advanced stages, leading to an apparent reverse-causal association between lower BMI and higher NPC mortality. 12 Further, there might exist some residual confounding, leading to Simpson's paradox, given that reported studies 13 typically adjusted for age, sex, smoking and alcohol history, cancer stage, and treatment information but not confounders such as education and occupations. Prior studies of NPC prognosis also may not have captured relevant measures of body size.
BMI is insufficient to thoroughly reflect body adiposity distribution. 14 Moreover, early-life body weight may affect disease outcomes throughout the life course, 15 yet research is lacking on the prognostic effects of BMI and body shape measured years prior to NPC diagnosis. Hence, using a prospectively followed patient cohort from our population-based case-control study of NPC in southern China, 16 we examined the effect of BMI and body shape at age 20 years, 10 years before diagnosis, and at diagnosis on overall and NPCspecific survival among NPC patients, with careful consideration to minimize reverse causation and residual confounding.

| Study population
The NPC cases included in our study comprised the patient cohort of our project entitled "NPC Genes, Environment, and EBV (NPCGEE)" in which cases were enrolled between 2010 and 2013 in Zhaoqing area of Guangdong Province, and Wuzhou and Guiping/Pingnan areas of Guangxi Autonomous Region in southern China. 16 Briefly, in the study base of 8 million people in these areas, we identified 3047 incident NPC cases during the study period; 2553 (83.8%) cases were included based on eligibility criteria designed to maximize the generability and validity of the results. Twenty-four cases were identified as not being incident NPC and excluded after reviewing medical records. Three cases without information on prediagnosis BMI and/or body shape were also excluded, leaving 2526 cases in the final analysis ( Figure S1).

| Ascertainment of exposures
Exposures were BMI and body shape at age 20 years, 10 years before diagnosis, and at diagnosis, all acquired from electronic lifestyle questionnaires completed by in-person interview between 2010 and 2013 (body shape and BMI at age 20 and 10 years before diagnosis) or medical record review between 2018 and 2022 (BMI at diagnosis).
Participants were interviewed as soon as possible after diagnosis, using a system of rapid case identification and study enrollment, with a median time of 2 days (interquartile range: 0-10 days) between diagnosis and interview. 16 BMI was calculated as weight in kilograms (kg) divided by the square of height in meters, and categorized into four groups: underweight (<18.5 kg/m 2 ), normal weight (18.5-22.9 kg/m 2 ), overweight (23.0-27.4 kg/m 2 ), and obese (≥27.5 kg/m 2 ) according to World Health Organisation guidelines for Asian populations. 17 Height at adulthood was assessed by height at age 20 years and regarded as stable. 18 Body shape was evaluated by the revised Stunkard's Figure Rating Scale, 19 which showed seven male and nine female visible schematic silhouettes, from extreme thinness to extreme adiposity (Method S1), and categorized into 1, 2, 3, 4 and 5-9 based on the distribution in the study population. Body shape 3 and normal BMI, categories with the largest sample size, were defined as reference groups.
Some adjacent groups (eg, overweight and obese) were combined for certain subgroup analyses due to insufficient sample size.

| Ascertainment of outcomes
All-cause mortality and NPC-specific mortality were the outcomes of interest. We used both measures because each of them has advantages and limitations. 20

| Covariates
Potential confounding variables were considered based on subject matter expertise, and included age at diagnosis; sex; residential area (Zhaoqing, Wuzhou, and Guiping/Pingnan); highest educational attainment (categorized into four groups based on Chinese education policy: illiterate/primary school, middle school, high school and vocational, and technical college/university and above); current occupation (farmer, blue collar, white collar, unemployed, and unknown/other); smoking history (never, former, and current); alcohol consumption (never, former, and current); treatment hospital (prefecture-level and university-affiliated/province-level); Karnofsky performance scale (KPS) before treatment (KPS ≥90 and <90); histological type (nonkeratinizing carcinoma and others); cancer stage at diagnosis; and treatment pattern. Cancer stage was restaged by study oncologists based on imaging reports according to the seventh version of the American Joint Committee on Cancer. 21 Treatment was categorized as concurrent chemoradiotherapy (CCRT), CCRT with either adjuvant chemotherapy (ACT) or induction chemotherapy (ICT), only radiotherapy, only chemotherapy, neither radiotherapy nor chemotherapy, and missing.

| Primary analyses
We used chi-squared tests for sex and residential area, Fisher's exact tests for other categorical variables, and Kruskal-Wallis tests for continuous variables to compare the distributions of NPC cases by BMI at age 20 years, 10 years before diagnosis, and at diagnosis.
Using time since diagnosis as the underlying timescale, we modeled all-cause mortality and NPC-specific mortality as follows. We used the Kaplan-Meier method to estimate cumulative survival proportions with 95% confidence intervals (CI). We used Cox proportional hazards regression to estimate (a) crude hazard ratios (HRs) for the associations between BMI/body shape and all-cause or NPCspecific mortality and (b) adjusted HRs controlling for continuous age at diagnosis, residential area, sex, educational attainment, occupation at recruitment, smoking history, alcohol consumption, and KPS before treatment, with stratification of baseline hazards by cancer stage.
Cases with missing BMI at diagnosis were classified into a separate category. Covariates were initially identified based on statistically significant univariate associations with all-cause and/or NPC-specific mortality (Table S1). We then calculated a variance inflation factor to check for multicollinearity, and omitted variables with a variance inflation factor >10 22 (eg, we omitted treatment pattern due to high collinearity with stage at diagnosis). We checked the proportional hazards assumption using Schoenfeld residuals and found that it was satisfied with P-values >.05, with the exception of cancer stage at diagnosis, which we therefore included as a strata variable. 23 P values for trend with increasing BMI or body shape were calculated by assigning ordinal values to categories.
To assess possible reverse causality, we estimated HRs stratified by cancer stage and compared nested models using likelihood ratio tests.

| Supplementary and sensitivity analysis
i. To evaluate associations with change in BMI from age 20 to 10 years before diagnosis, and to diagnosis, we first used linear regression to calculate the BMI slope across the three time points as the outcome, with age as the exposure. We then used the slope as the exposure in Cox regression models with restricted cubic   Figure 1A for BMI; P = .782, Figure 1D for body shape). In contrast, higher BMI 10 years before diagnosis (P = .033, Figure 1B) and at diagnosis (P < .001, Figure 1C) was associated with lower all-cause mortality. Similarly, larger body shape 10 years before diagnosis (P = .042; Figure 1E) and at diagnosis (P < .001; Figure 1F) was significantly associated with lower all-cause mortality. Associations of BMI and body shape with cumulative NPCspecific survival followed the same patterns ( Figure 1G-L).
Crude and adjusted HRs from Cox proportional hazards regression models are shown in Patterns of association with NPC-specific mortality were similar to those for all-cause mortality (Tables 2 and 3). BMI and body shape at age 20 years were not associated with allcause or NPC-specific mortality ( Table 2). Being overweight or obese 10 years before diagnosis was associated with lower all-cause and NPC-specific mortality, but the associations were diminished in adjusted models (Table 2). Likewise, thinner body shape (especially body shape 2) 10 years before diagnosis was associated with greater all-cause and NPC-specific mortality than body shape 3, but associations were weaker than those for body shape at diagnosis ( Table 2).

| Supplementary and sensitivity analysis
(i) Increasing BMI over time was linked with lower all-cause and NPCspecific mortality up to a threshold of below approximately 0.25 kg/m 2 F I G U R E 1 Kaplan-Meier curves for overall survival probabilities by BMI at age 20 years (A), 10 years before diagnosis (B), and at diagnosis (C) and body shape at age 20 years (D), 10 years before diagnosis (E), and at diagnosis (F); NPC-specific survival probabilities across BMI at age 20 years (G), 10 years before diagnosis (H), and at diagnosis (I) and body shape at age 20 years (J), 10 years before diagnosis (K), and at diagnosis (L). Log-rank tests were used to compare differences among groups.
T A B L E 2 Hazard ratios for body mass index and body shape in association with all-cause and NPC-specific mortality among 2526 nasopharyngeal carcinoma cases during 2010-2013 in southern China. per year, after which the protective association gradually decreased,  (Table S4). No effect modification was found by smoking history (Table S5). (iv) Pooled HRs using multiple imputation to replace missing/unknown values were similar to the main results, but no statistically significant associations were observed with BMI or body shape 10 years before diagnosis (Table S6). (v) Complete-case analysis yielded results closely resembling the main findings (Table S7).

| Interpretation of key findings
In this population-based prospective cohort study of NPC in southern China, we investigated the effect of BMI and body shape at age 20 years, 10 years before diagnosis, and at diagnosis on allcause and NPC-specific mortality, taking particular care to address concerns of reverse causation. We found that being overweight or obese at diagnosis conferred approximately 25% lower mortality than normal weight after an average of 5.5 years of follow-up, whereas the thinnest body shape conferred 68% higher mortality than normal body shape. The associations with BMI and body shape at diagnosis were not modified by cancer stage, and indeed the positive associations with lower BMI and thinner body shape were stronger for early-stage NPC patients. The lack of heterogeneity by stage at diagnosis, combined with the detection of similar, albeit somewhat weaker, associations with BMI and body shape 10 years before diagnosis, suggest that the results were not likely to be due mainly to reverse causation. BMI and body shape at age 20 years did not correlate with overall or cause-specific mortality.

| Comparisons to previous literature
Previous investigators reported generally inverse associations between BMI and NPC mortality 6,9,10,[24][25][26][27]  With thorough adjustment for potential confounders, stratification by cancer stage, and evaluation of associations with anthropometrics 10 years prior to diagnosis, we found stable associations between BMI or body shape and NPC prognosis. An inverse association of BMI with breast cancer mortality has previously been validated through Mendelian randomization analysis, 29,30 supporting the biological plausibility of a prognostic effect of body size on cancer survival.
We also found that lean body shape at diagnosis and, to a lesser extent, 10 years earlier was significantly associated with increased NPC mortality. Thus far, to our knowledge, no research has investigated associations between body shape and NPC prognosis; however, our findings are consistent with prior results for BMI, which can be correlated with body shape. 7,31 To our knowledge, there is no previous research investigating the association between BMI at age 20 years and overall mortality in NPC patients. Although we did not observe a significant association between BMI at age 20 years and all-cause mortality overall for NPC patients, we found that among women, being overweight/obese was associated with 39% lower mortality. Obesity at age 20 years, which may represent early-life overnutrition and a sedentary lifestyle, was associated with higher mortality risk. 32 However, the impact of earlylife BMI might vary by cancer site, 29 sex, 33 and/or culture/ethnicity. Alternatively, our findings may have been distorted by misclassification, since BMI at age 20 years was self-reported based on recollection, although exposure information was collected prospectively relative to mortality outcomes.
Few studies have investigated the prognostic impact of BMI change from at age 20 years to diagnosis, whether for NPC or other cancer types. Our results showed that increasing BMI was linked with lower mortality until a certain threshold, after which the apparent protective effect gradually decreased. These findings suggest that excess weight gain may have an adverse prognostic effect, but our study lacked sufficient power at the extremes of BMI, body size, and weight change to detect any such patterns.
Our results can be interpreted from the perspectives of biology and clinical practice. The primary therapy for NPC is radiation-based, with or without chemotherapy. Radiation side effects on the oral and pharyngeal mucosa are often painful, making patients reluctant to eat, resulting in a negative nutrition balance. The average radiation course for NPC patients is approximately 45 days. Those with poorer nutrition at baseline, as indicated by lower BMI and thinner body shape, could be less tolerant of radiotherapy side effects, resulting in treatment discontinuation and poorer survival outcomes.

| Strengths and limitations of studies
The study has several strengths. First, it is a population-based study with low loss to follow-up, thereby maximizing generalizability in southern China, where NPC is endemic, and minimizing potential selection bias. Second, information on BMI and body shape was collected prospectively, prior to follow-up for vital status, thereby minimizing potential recall bias. Third, we were able to stratify by cancer stage to investigate reverse causation. Fourth, we adjusted for many covariates, including smoking, alcohol, education, occupation, and others, thereby reducing potential bias from confounding. Finally, this is the first research study investigating the prognostic impact of BMI and body shape at age 20 years and 10 years before diagnosis of NPC, making our results novel and lending insight into how factors preceding NPC onset by a decade may still influence disease outcomes.
There are several limitations in our research. First, body size and shape at age 20 years and 10 years before diagnosis were assessed by self-report, and are therefore prone to error, although this error is likely to be nondifferential due to the prospective data collection.
Second, despite the large size of our patient cohort, numbers of deaths were limited among patients in the lowest and highest categories of BMI, body shape, and weight change, thereby reducing the power to detect true trends and differences between groups. Third, some patients had missing values for BMI at diagnosis, cancer stage, KPS, and treatment hospital, potentially creating bias. Nevertheless, the pooled HRs after multiple imputation were consistent with the primary results, pointing to the robustness of our findings. Fourth, we could not obtain additional information on potentially relevant anthropometric measures, such as muscle, body fat, and fat distribution, which would enable further characterization of body composition (eg, have high BMI due to high muscle density, as opposed to high fat tissue). In addition, our results for certain BMI categories may not be generalizable to other populations with different criteria for defining overweight according to BMI, although the significant dose-response relationship when BMI was treated as a continuous variable allays somewhat a concern. Finally, we lacked detailed information on treatment course, preventing us from determining whether lower-BMI and thinner patients were more likely to discontinue radiotherapy.

| CONCLUSION
In summary, we found that being overweight or obese at diagnosis was associated with decreased all-cause and cause-specific mortality among NPC patients, whereas having a leaner body shape was associated with increased mortality, compared to normal weight/body shape. The lack of heterogeneity of these associations by cancer stage at diagnosis, as well as their persistence 10 years prior to diagnosis, suggests that they are not likely to be attributable mainly to reverse causation. If, as we hypothesize, these associations are due to poorer radiotherapy adherence among NPC patients with lower BMI and thinner body shape, then greater clinical attention may be needed to ensure better nutrition and treatment tolerance for improved survival outcomes in this group of patients.

AUTHOR CONTRIBUTIONS
The original study was initiated by Weimin Ye, Hans-Olov Adami, and