Pre-diagnostic body mass index and weight change in relation to colorectal cancer survival among incident cases from a population-based cohort study

The Norwegian Counties Study is a population-based Norwegian cohort study described in detail elsewhere [23, 24]. In short, participants were examined by a team of trained nurses at health screenings up to three times between 1974 and 1988. The attendance rate was >80 % at all three screenings, and 92,234 men and women attended at least one screening. At each screening, the participants’ height was measured to the nearest centimeter and weight to the nearest 0.5 kilo. Information on lifestyle factors such as smoking habits and recreational and occupational physical activity during the last year was collected with a questionnaire.

Using the unique personal identification number assigned to all Norwegian citizens, information on each participant’s education was obtained from records of the censuses in 1970, 1980, 1990, and 2001. The most recent information available was used.

CRC cases were identified through linkage with the Cancer Registry of Norway, i.e. cancers coded as 153 or 154 according to the International Classification of Diseases, Seventh edition (ICD-7). The cases were categorized as colon cancer (ICD-7: 153) or rectal cancer (ICD-7: 154). Furthermore, cancers of the appendix, cecum, ascending or transverse colon (including the hepatic and splenic flexures) (153.0, 153.1, and 153.6) were categorized as proximal colon cancer. Cancers that occurred in the descending colon, sigmoid colon, or rectosigmoid junction (153.2 − 153.4) were categorized as distal colon cancer. We only considered the first cancer diagnosis. Participants diagnosed with multiple malignant tumors at the date of first diagnosis were included as CRC cases if all the cancers occurred in the colorectum. Correspondingly, multiple cancers were included as a cancer of one of the subsites (colon, proximal colon, distal colon, or rectum) if they all occurred in the same subsite. The cases were classified according to stage at diagnosis as localized, regional, or distant. For participants with multiple malignant tumors, stage at diagnosis was defined as the stage of the most advanced.

We identified 2786 cases of CRC among the cohort participants. Of these, 69 were diagnosed with multiple malignant tumors.

For each case, we obtained information on weight, smoking, and physical activity level from the most recent screening. This screening was defined as the case’s exposure screening if 1) no information was missing and 2) the screening took place at least 3 years prior to diagnosis. Otherwise, the most recent preceding screening fulfilling these requirements was defined as the exposure screening. Pre-diagnostic BMI (kg/m²) was calculated as weight at the exposure screening divided by height squared, using the mean height from all attended screenings. We excluded 54 cases with missing information on weight, smoking, and physical activity at all three screenings, 35 cases diagnosed before information was collected, and 44 cases diagnosed <3 years after the information was collected. We furthermore excluded 5 cases with no height measurements and 6 cases without information on education. We also excluded 23 cases with BMI <18.5 kg/m² at the exposure screening due to concerns that their weight was influenced by pre-existing CRC or other underlying health problems that may influence survival. Finally, we excluded 97 cases with unknown stage at diagnosis, 2 cases with date of diagnosis equal to date of death, and 4 cases with date of emigration prior to their date of diagnosis.

In total, 2516 CRC cases, 1336 men and 1180 women, diagnosed between 1978 and 2012, were included in analyses of BMI and CRC survival (Additional file 1 : Figure S1). In men, 847 of the cases were colon cancers, of which 443 occurred in the proximal and 369 in the distal colon, and 478 cases were rectal cancers. In women, 808 cases were colon cancers, 493 were proximal colon cancers, 294 were distal colon cancers, and 367 were rectal cancers.

For analyses of weight change, we furthermore excluded 622 cases with no weight measurements prior to the exposure screening, leaving 999 men (637 colon cancers, 339 proximal colon cancers, 270 distal colon cancers, 351 rectal cancers) and 895 women (608 colon cancers, 383 proximal colon cancers, 207 distal colon cancers, 283 rectal cancers) (Additional file 2: Figure S2).

We determined the power of detecting an effect estimate per 5 unit increase in BMI of 1.20 with a significance level of 0.05 in our study sample [25]. We used the standard deviation of BMI among CRC cases, 3.2 kg/m² in men and 4.5 kg/m² in women. In men, the power was 0.87 for CRC, 0.69 for colon cancer, 0.41 for proximal colon cancer, 0.31 for distal colon cancer, and 0.46 for rectal cancer. In women, the power was 0.97, 0.87, 0.68, 0.45, and 0.60 for CRC, colon cancer, proximal colon cancer, distal colon cancer, and rectal cancer, respectively.

The cases were followed from the date of diagnosis until death, emigration, or December 31, 2012. Information on death and emigration was obtained from the Cause of Death Registry and the National Population Registry. CRC-related death was the primary outcome and death from all causes was the secondary outcome. Deaths with cause 153–154 (ICD-8 and ICD-9) or C18–21 (ICD-10) were considered CRC-related deaths.

Pre-diagnostic BMI was categorized according to WHO guidelines: 18.5 − 24.9 kg/m² (normal weight), 25 − 29.9 kg/m² (overweight), and ≥30 kg/m² (obese) [26]. Weight change was defined as weight at the exposure screening minus initial weight, i.e., weight at the first attended screening, and was categorized as ≤−2.0 kg (weight loss), −1.9 − 1.9 kg (weight maintenance), 2.0–7.9 kg (moderate weight gain), and ≥8.0 kg (large weight gain).

We examined BMI and weight change in relation to risk of death from CRC, our primary outcome, and risk of death from all causes. Hazard ratios (HRs) and 95 % confidence intervals (CIs) were estimated with Cox regression. We used a stratified Cox model with stage of disease at diagnosis as the strata (localized, regional, or distant) and time since diagnosis as the time variable. We adjusted for smoking (never, former, current) and level of recreational and occupational physical activity combined (sedentary, moderately active, or active) [23], both measured at the exposure screening, education (primary schooling (≤9 years), secondary education (10–12 years), or university level education (≥ 13 years)), age at diagnosis (continuous) and year of diagnosis (< 1990, 1990–94, 1995–99, 2000–2004, and ≥2005). Analyses of weight change were also adjusted for initial BMI (continuous). Adjustment for time between the exposure screening and diagnosis did not change the results and was not included in the model. To test for trend, BMI/weight change was modeled continuously and evaluated with a Wald test. All analyses with continuous weight change were restricted to subjects who maintained or gained weight because we hypothesized that weight loss may cause a J or U-shaped relation. We furthermore tested whether the associations with CRC-specific mortality differed in the colon and rectum or in the proximal and distal colon by testing the difference in trends with a Wald statistic. To assess whether associations differed with sex, we included interaction terms between sex and BMI (continuous) or between sex and weight gain (continuous). For CRC, we also examined whether the association between weight gain (continuous) and mortality differed with initial BMI (<25 kg/m² and ≥25 kg/m²). Tests of interaction were evaluated with a likelihood ratio test and were done for CRC-specific mortality only. Proportional hazards assumptions were evaluated with a test based on the Schoenfeld residuals. The tests did not indicate violations of the assumption. All tests were two sided, and P < 0.05 was considered statistically significant. The analyses were performed with SAS version 9.4 (SAS Institute, Cary, NC).

Mean age at diagnosis in the BMI study sample (n = 2516) was 66.4 years (range 32.0–85.6) in men and 66.8 years (range 36.0–86.4) in women. During a mean follow-up of 5.8 years, 703 men and 543 women died. Of these, 507 men and 432 women died from CRC. Time from BMI measurement to diagnosis ranged from 3.0 to 37.5 years (mean 17.8 years). Mean pre-diagnostic BMI was similar in men and women, 26.0 kg/m² and 25.7 kg/m², respectively. For both men and women, the proportion with a university education was highest among those with normal weight (Table 1). With increasing BMI, the proportion of sedentary subjects increased, the proportion of current smokers decreased, and mean systolic blood pressure increased. Furthermore, age at diagnosis increased and the proportion with distant stage disease decreased with increasing BMI in women.

In men with proximal colon cancer, pre-diagnostic obesity was associated with higher risk of CRC-related death than normal weight, HR = 1.85 (95 % CI 1.08–3.16) (Table 2). BMI was not significantly associated with CRC-specific mortality in men with CRC, colon cancer, distal colon cancer, or rectal cancer. In women, we found no significant associations between BMI and risk of CRC-related death after diagnosis of CRC or cancer of any colon subsite. However, obese women had significantly increased risk of CRC-related death after diagnosis of rectal cancer, HR = 1.93 (95 % CI 1.13–3.30). We found no significant interactions between BMI and sex (P_interaction ≥ 0.26). The associations between BMI and risk of CRC-related death did not differ significantly with subsite: colon vs. rectum (P_{heterogeneity} ≥ 0.27), proximal vs. distal colon (P_{heterogeneity} ≥ 0.33). In contrast with the results on CRC-related death, pre-diagnostic obesity was associated with higher all-cause mortality in men with CRC and colon cancer (Additional file 3: Table S1). In women, associations between pre-diagnostic BMI and all-cause mortality were similar to the associations found for CRC-specific mortality.

In the weight change study sample (n = 1894), 533 men and 414 women died during follow-up. Of these, 372 men and 326 women died from CRC. Mean weight change was 2.5 kg in men and 2.0 kg in women. Cases who lost weight prior to diagnosis had a higher initial BMI and were more likely to be current smokers than cases who maintained or gained weight (Table 3). Furthermore, the highest proportion of sedentary subjects was found among those who gained ≥8 kg. In men, the proportion with distant stage disease was lowest among those who had maintained their weight and the proportion increased with increasing weight gain, whereas in women the proportion with distant stage disease decreased with increasing weight gain.

Pre-diagnostic weight gain was not significantly related to CRC-specific mortality in men, but weight loss was associated with significantly increased risk of CRC-related death compared to weight maintenance in men with rectal cancer, HR = 1.78 (95 % CI 1.06–3.00) (Table 4). In women, weight gain was significantly associated with increased risk of death from CRC after diagnosis of CRC, HR = 1.18 (95 % CI 1.01–1.37) per 5 kg weight gain, P_trend = 0.03. Moreover, in women with colon cancer, moderate and large weight gain were both associated with significantly increased risk of CRC-related death compared to weight maintenance, HRs = 1.54 (95 % CI 1.08–2.20) and 1.64 (95 % CI 1.03–2.63), respectively, P_trend = 0.03. For proximal colon cancer, we observed increased risk in women with moderate weight gain, HR = 1.59 (95 % CI 1.01–2.49), but not in women with large weight gain. In women with distal colon cancer, large weight gain was associated with significant, increased risk of CRC-related death compared to weight maintenance, HR = 2.99 (95 % CI 1.27–7.05), P_trend = 0.04. Weight change was not associated with CRC-specific mortality in women with rectal cancer. No significant interactions between weight gain and sex were observed (P_interaction ≥ 0.18). The associations did not differ significantly with subsite; colon vs. rectum (P_{heterogeneity} ≥ 0.38), proximal vs. distal colon (P_{heterogeneity} ≥ 0.12). We did not observe significant interactions between BMI and weight gain (P_interaction ≥ 0.33). The results on all-cause mortality were similar to the results on CRC-specific mortality, except that weight gain was not significantly associated with higher all-cause mortality in women with CRC overall (Additional file 4: Table S2).