Table 2.
Estimated change points in the association between BMI and mortality among never-smokers, and associations with mortality below and above the change point, from piecewise two-line models for the 5-year post-BMI exclusion period
| BMI change point, kg/m2 (95% CI) | HR per 5 kg/m2 BMI increase below change point*(95% CI) | HR per 5 kg/m2 BMI increase above change point (95% CI) | ||
|---|---|---|---|---|
| All-cause mortality | 25 (25–25) | 0·81 (0·80–0·82) | 1·21 (1·20–1·22) | |
| Level 1 outcomes | ||||
| Communicable diseases | 26 (26–26) | 0·73 (0·71–0·76) | 1·28 (1·24–1·31) | |
| Non-communicable diseases | 25 (25–25) | 0·83 (0·81–0·84) | 1·22 (1·21–1·23) | |
| Injuries and external causes | 27 (26–28) | 0·75 (0·71–0·80) | 1·10 (1·04–1·17) | |
| Level 2 outcomes (ICD-10 chapters/codes) | ||||
| Cancers (C) | 21 (20–25) | 0·88 (0·80–0·97) | 1·13 (1·12–1·14) | |
| Blood and endocrine (D50–89, E) | 22 (22–29) | 0·43 (0·35–0·54) | 1·42 (1·37–1·48) | |
| Mental and behavioural (F) | 24 (21–25) | 0·31 (0·22–0·44) | 1·05 (0·86–1·27) | |
| Neurological (G) | 26 (25–27) | 0·68 (0·66–0·70) | 0·98 (0·96–1·01) | |
| Cardiovascular (I) | 25 (25–25) | 0·89 (0·87–0·91) | 1·29 (1·27–1·30) | |
| Respiratory (J23–99) | 25 (24–25) | 0·53 (0·50–0·56) | 1·25 (1·21–1·29) | |
| Liver cirrhosis (K70·3/71·7/74·3–6) | 23 (22–27) | 0·75 (0·48–1·16) | 1·44 (1·33–1·55) | |
| Digestive (K, excluding cirrhosis) | 24 (22–25) | 0·79 (0·72–0·86) | 1·32 (1·28–1·36) | |
| Musculoskeletal (M) | 24 (24–25) | 0·45 (0·39–0·53) | 1·23 (1·15–1·32) | |
| Urogenital (N) | 25 (24–25) | 0·84 (0·77–0·93) | 1·45 (1·39–1·51) | |
| Accident, transport-related (V) | NA* | 1·00 (0·90–1·11) | .. | |
| Accident, excluding transport (W/X00–59) | 27 (26–28) | 0·71 (0·66–0·77) | 1·17 (1·09–1·26) | |
| Self-harm and interpersonal violence (X60–Y09) | NA* | 0·87 (0·80–0·94) | .. | |
HR=hazard ratio. ICD-10=International Classification of Diseases, 10th revision. NA=not available.
*
For transport-related accidents, and self-harm and interpersonal violence, there was little or no evidence against linearity (figure 2) so a single linear effect without change point was estimated.