A CMR of 73 per 10,000 pys translates to 73 deaths occurring among 10,000 people over a one-year period or to 73 deaths occurring among 20,000 people over a 6-month period. Standardised Mortality Ratio (SMR): describes the extent to which mortality in a cohort differs from that which would be seen in an ‘average’ population, matched for age and gender. An SMR of 5.7 means that there were 5.7 times more deaths occurring in the cohort PD0325901 cell line than would have

occurred in a sample of the general population who had the same distribution of age and gender. Crude mortality rates (CMR) per 10,000 person-years (pys) were calculated for all-cause mortality, and drug-related poisoning deaths. An individual’s risk period began at the date of their earliest observation

in the cohort on or after 1st April 2005 and ceased at the end of data collection (31st March, 2009) or the date of death, if earlier. Individuals already in treatment on 1st April, 2005 began their time at risk from that date. Observed deaths (O) were compared to gender and age appropriate expected mortality (E) to derive standardised mortality ratios (SMR = O/E). The expected mortality was calculated by multiplying the (disease specific) mortality rate observed in the general population by the person years of follow-up seen in the analysis cohort, Screening Library clinical trial matched by age and gender (indirect method). Confidence intervals for CMRs and SMRs were calculated using a normal approximation to the Poisson distribution for the observed number of deaths. All p values are two sided. Following strong prior information for drug-related poisonings

(King et al., 2012 and King Terminal deoxynucleotidyl transferase et al., 2013), we assessed whether mortality differences between males and females persist with age by testing for an interaction between gender and age-group. The interaction was evaluated by testing whether the relative risk, comparing the drug-related poisoning mortality rate between males and females, was equal across age groups, using a chi-squared test. Evidence for presence of an interaction was set at p < 0.01. As a sensitivity analysis, we assessed whether evidence for an age and gender interaction was due to differences in behavioural risk factors by carrying out an adjusted analysis on the subcohort of treated individuals for whom we have information available on risk factors (n = 151,983). This is described in the supplementary material 1. Cause-specific CMRs and SMRs were first calculated at the ICD-10 Chapter level. To retain statistical power the a priori analysis strategy was to present CMR/SMRs at subsequent ICD-10 lower, more detailed descriptive levels if (a) the SMR for the higher level was ≥5; and (b) the expected number of deaths for the lower level was ≥5.