As previously reported, the primary endpoint was decreased by 37% (95% CI 17–53%) on active treatment compared with placebo, acute coronary events by 36% (9–55%) and stroke by 48% (11–69%; all p < 0.05) . All-cause mortality declined by 27% (−1.48%), which was of borderline significance p = 0.0592) overall, but significant in the final year .
Change in lipids, lipoproteins and apolipoproteins
At baseline, patients allocated to active atorvastatin or placebo were closely matched to their lipid, lipoprotein and apolipoprotein levels as previously reported . Thus, in the 2,627 patients in whom complete laboratory data were available at baseline, mean serum cholesterol was 5.36 ± 0.82 mmol/l (mean ± SD), LDLC 3.04 ± 0.71 mmol/l, non-HDLC 3.95 ± 0.83 mmol/l, HDLC 1.40 ± 0.33 mmol/l, ApoA-I 1.53 ± 0.29 g/l and ApoB 1.16 ± 0.24 g/l. The mean difference in LDLC concentration on atorvastatin compared with placebo throughout the trial was 40.9% (mean 95% CI 40.1–41.6%), whereas ApoB decreased by 24.3% (23.4–25.2%; both p < 0.0001). The decrease on atorvastatin compared with placebo in ApoB:ApoA-I was 27.2% (−28.5, −25.9%; p < 0.0001) and, in LDLC:HDLC, −44.6% (−45.6, −43.5%; p < 0.001). HDLC was significantly increased on active treatment compared with placebo by 1.6% (1.0–2.1%; p < 0.05), but ApoA-I was not significantly changed by atorvastatin.
Baseline lipoproteins and cardiovascular outcomes
Table 1 shows the change in HR adjusted by treatment for a primary endpoint in the combined treatment groups predicted by a 1 SD difference in various lipoprotein variables and their ratios. These are ranked according to p values by χ
2. These are inversely related to the strength of the association. The most statistically significant increments in HR were predicted by the ApoB:ApoA-I ratio. Although the less significant increment in HR for a 1 SD change in LDLC:HDLC or LDLC was greater than for ApoB:ApoA-I, this was less statistically significant because the SD was smaller for ApoB:ApoA-I. The level of statistical significance which is related to the Cox regression coefficient (β) and its standard error reveals the true strength of the relationships.
Table 2 shows the subgroup analysis by treatments comparing the predictability of the risk of primary endpoints among studied lipids and lipoproteins. For the placebo group, ApoB:ApoA-I ratio remained the strongest significant risk factor for primary endpoint. Compared with both treatment groups combined, the rank orders among the first four lipid/lipoproteins remained the same with only slight differences in the last four lipids. For the atorvastatin group, all studied lipid/lipoproteins variables were insignificant.
Table 3 shows the intertertiles analysis of ApoB:ApoA-I ratio at baseline and HR, both unadjusted for treatment effect (model 1) and adjusted for treatment effect (model 2). The results based on ‘adjusted for treatment effect’ and HR for primary endpoints, CHD endpoints, total mortality and stroke were very similar to those from an unadjusted analysis. The relationship between both cardiovascular risk and all-cause mortality and tertiles of the ApoB:ApoA-I ratio was stronger compared with other lipoprotein variables (Fig. 1a,b). Furthermore, it was strongest for the CHD component of cardiovascular risk (Fig. 1c) and weakest for stroke for which it had no statistical significance (Fig. 1d). Both the ApoB:ApoA-I and the LDLC:HDLC ratios also identified patients at greater risk of CHD than non-HDLC:HDLC as judged by their ROCs (p = 0.0005 and p = 0.0125 respectively; Fig. 2). However, the apparently greater AUC for ApoB:ApoA-I as opposed to LDL:HDLC did not achieve statistical significance.
Lipoprotein levels on treatment and cardiovascular outcome
The first time apolipoproteins were measured per protocol after baseline in the participants as a whole was at 1 year after commencing treatment. By then, 60 primary endpoints had occurred. Attention to this early benefit has previously been drawn . It meant that many of the total number events in the trial had occurred before the changes in apolipoproteins with the treatment were measured for the first time. Thus, it was not possible to relate individual treatment changes to outcomes, because of insufficient subsequent events. However, the time-dependent Cox proportional hazard regression model, which included percentage change in ApoB:ApoA-I adjusted for treatment effect as a covariate, predicted that the 27.2% decrease in ApoB:ApoA-I on active treatment would produce a 30.2% (95% CI 4.1–49.2%) decrease in the primary endpoints, which is similar to the 37% decrease observed in the full study. The reduction in primary endpoints predicted by the decrease of ApoB:ApoA-I ratio was almost exclusively due to CHD events. Thus, based on a similar time-dependent Cox proportional hazard regression model, it was estimated that the same 27.2% decrease in ApoB:ApoA-I ratio on active treatment would produce a 32% (5.4–51.2%) reduction in CHD risk which was close to the 36% decrease observed. None of the lipid, lipoprotein or apolipoprotein variables, or their ratios, forecast the 48% decrease in the incidence of strokes.