The Medical Research Council National Survey of Health and Development (NSHD) is a socially stratified cohort of 5362 individuals followed up since birth in England, Scotland and Wales in March 1946. The sampling procedure and follow-up have been described in detail elsewhere[19
]. 3322 (62.0%) were interviewed at age 36. Of the remaining 2040, 323 (6.0%) had died, 649 (12.1%) had emigrated, 510 (9.5%) had previously refused to participate and 558 (10.4%) were untraced.
At age 36 research nurses administered the Present State Examination (PSE) to 3293 participants at home[20
]. The PSE assesses the frequency and severity of psychiatric symptoms in the preceding month and can be coded as an index of definition (PSE-ID) which ranges from 1-7. We distinguish three groups: those with an ID of 3 or more (22.0%) who typically have at least 5 psychiatric symptoms and were likely to have a psychiatric disorder; those with an ID of 2 (30.7%), who were mildly symptomatic with 1-4 symptoms; and those with a PSE-ID of 1 who had no symptoms (47.2%)[21
]. Using the CATEGO computer-based classification system, we determined the presence of anxiety and depression disorders across PSE-ID categories. The highest PSE-ID category (PSE-ID= 3 or more) contained all cases of co-morbid depression and anxiety (n = 37), all cases of depression alone (n = 172), and the majority of cases of anxiety alone (n = 362). The remaining third of anxiety cases (without co-morbid depression) (n = 148) were present in PSE-ID category 2. No cases of anxiety or depression were present in PSE-ID category 1.
3283 of the 3293 men and women with a PSE-ID score were identified on the National Health Service Central Register and we were notified of all deaths. The underlying cause of death was coded according to ICD-9 or ICD-10. Deaths from "externalizing" disorders including violent accidental and suicidal deaths (ICD9 codes 291 - 292, 295 - 305, 307 - 309, 311 - 316, 570 - 571.3, 800 - 994, 1800 - 1869, 1880 - 1999 and ICD10 codes F61 - F69, K70 - K71, S00 - X99, Y85 - Y98) were identified.
Factors previously found to be associated in this cohort with premature adult mortality[22
] were chosen as potential confounders. They included social class of origin, based on father's occupation at 4 years, and own social class at age 26 years, according to the Registrar General's 1971 classification, dichotomized to manual and non-manual. Childhood cognitive ability was assessed at age 15 using the Heim AH4 test,[25
] the Watts Vernon reading test,[26
] and a mathematics test; scores were standardised, summed, standardised again then categorised into fourths. Highest educational qualifications at age 26 were categorised into three groups: no qualifications; Ordinary levels (usually taken at 16); and Advanced levels (usually taken at 18 for university entry) or above.
Potential mediating variables were smoking behaviour, alcohol consumption and physical health status. Current smokers, ex-smokers and lifelong non-smokers were distinguished from information collected at adult follow-ups to age 36 years. Alcohol consumption was recorded at 36 years using a five day diary; total grams per day were calculated and categorised into fifths. Physical health status at age 36 has previously been assessed in detail[27
]. Cohort members were categorised into those in the best (10.3%), intermediate (62.8%) or in worst physical health (26.9%), on the basis of measured blood pressure, lung function and body weight, self reported health problems and disability, and recent hospital admissions.
We used survival curves, obtained by the Kaplan-Meier method, to compare the cumulative death rates between 36 and 60 years for those with and without psychiatric disorder. Cox's proportional hazards models were used to investigate the relationship between psychiatric disorder and adult mortality rates. The proportional hazards assumption was checked using the spthtest function in Stata. Follow-up time (in months) was from the cohort's 36th birthday until the first of death, emigration, or the end of March 2006. If death had not occurred, follow-up was treated as censored. Sex adjusted hazard ratios (HRs) for psychiatric disorder at 36 years were then further adjusted, in turn, for potential confounders and mediators. A further model included all variables. In these analyses, those with missing data on any potential confounder or mediator were assigned to a separate group. Sensitivity analyses were undertaken to compare the effects of psychiatric disorder on adult mortality risk, first including and then excluding this missing category.
We identified 22 cases of schizophrenia in the sample. All analyses were repeated, censoring for cases of schizophrenia and also censoring for deaths from 'external' causes (including accidental deaths and suicide). Analyses were performed using Stata 10.0[28
]. All results presented have been weighted to adjust for the social class stratification in the original sample.