Childhood intelligence in relation to major causes of death in 68 year follow-up: prospective population study

Objectives To examine the association between intelligence measured in childhood and leading causes of death in men and women over the life course. Design Prospective cohort study based on a whole population of participants born in Scotland in 1936 and linked to mortality data across 68 years of follow-up. Setting Scotland. Participants 33 536 men and 32 229 women who were participants in the Scottish Mental Survey of 1947 (SMS1947) and who could be linked to cause of death data up to December 2015. Main outcome measures Cause specific mortality, including from coronary heart disease, stroke, specific cancer types, respiratory disease, digestive disease, external causes, and dementia. Results Childhood intelligence was inversely associated with all major causes of death. The age and sex adjusted hazard ratios (and 95% confidence intervals) per 1 SD (about 15 points) advantage in intelligence test score were strongest for respiratory disease (0.72, 0.70 to 0.74), coronary heart disease (0.75, 0.73 to 0.77), and stroke (0.76, 0.73 to 0.79). Other notable associations (all P<0.001) were observed for deaths from injury (0.81, 0.75 to 0.86), smoking related cancers (0.82, 0.80 to 0.84), digestive disease (0.82, 0.79 to 0.86), and dementia (0.84, 0.78 to 0.90). Weak associations were apparent for suicide (0.87, 0.74 to 1.02) and deaths from cancer not related to smoking (0.96, 0.93 to 1.00), and their confidence intervals included unity. There was a suggestion that childhood intelligence was somewhat more strongly related to coronary heart disease, smoking related cancers, respiratory disease, and dementia in women than men (P value for interactions <0.001, 0.02, <0.001, and 0.02, respectively).Childhood intelligence was related to selected cancer presentations, including lung (0.75, 0.72 to 0.77), stomach (0.77, 0.69 to 0.85), bladder (0.81, 0.71 to 0.91), oesophageal (0.85, 0.78 to 0.94), liver (0.85, 0.74 to 0.97), colorectal (0.89, 0.83 to 0.95), and haematopoietic (0.91, 0.83 to 0.98). Sensitivity analyses on a representative subsample of the cohort observed only small attenuation of the estimated effect of intelligence (by 10-26%) after adjustment for potential confounders, including three indicators of childhood socioeconomic status. In a replication sample from Scotland, in a similar birth year cohort and follow-up period, smoking and adult socioeconomic status partially attenuated (by 16-58%) the association of intelligence with outcome rates. Conclusions In a whole national population year of birth cohort followed over the life course from age 11 to age 79, higher scores on a well validated childhood intelligence test were associated with lower risk of mortality ascribed to coronary heart disease and stroke, cancers related to smoking (particularly lung and stomach), respiratory diseases, digestive diseases, injury, and dementia.

Matching death records to individuals in Scottish Mental Survey 1947

Matching death records without ID numbers to individuals in the Scottish Mental Survey 1947 database
Scotland-recorded deaths between 1 July 2015 and 31 December 2015 were provided to the study without unique ID numbers for individuals. Therefore an electronic matching program was run to amalgamate individuals' death data with their respective childhood data. The program used key variables (first name, surname, date of birth) to match individuals in the two datasets, utilising Python programming language (https://www.python.org/), the NumPy library for scientific computing (http://www.numpy.org/), and, the pandas library for data analysis (http://pandas.pydata.org/). In attempting to match 2453 individuals' death data with their childhood records, the program successfully matched 1939 individuals, whereas 514 failed, due largely to first name variants between the two files. Of the failed matches, 315 were resolved manually, leaving 199 non-matches that were excluded from analyses.

Reconciling mismatches on sex and date of birth between individuals' records in the mortality death and the Scottish Mental Survey 1947 data
Following the merging of files, there were numbers of mismatches on sex and date of birth identified between the historical records and the linkage data. There were 296 mismatches on sex and 3114 mismatches on date on birth. Manual checks on identifiable data were carried out to reconcile these differences (details below) resulting in 987 exclusions due to mismatching (178 of these also met other exclusion criteria).
Date of birth mismatches -There were 3114 mismatches on date of birth (4.3% of the traced sample), which is a similar proportion of mismatches to a previous linkage study in Scotland (Hart et al., Public Health, 2003). The table below summarises the differences in dates of birth between the two records. Firstly, we accepted mismatches as correct matches if there was a difference of 9 days or less in the dates, or, if there was a difference of 10 days with only one digit different (n=1941). Secondly, we conducted manual checks on samples of mismatches, and confirmed a match if at least one of the following criteria were met: (1) where the name that matched was unusual; (2) where there was an unusual middle name that matched, (3) when there was a confirmed maiden name. This process of manual checking found correct matches for all mismatches where the different dates were exactly one month different, and, where day and month appeared in reverse to the other, and so we accepted these as true matches. Sex mismatches -Participants' names were first checked to ensure compatibility between records, and then the forename helped to determine sex. Where there were name ambiguities on sex (n=9), AP, AT, CMC, and IJD, reached a consensus. One match was excluded from analyses due to a mismatch on both forename and date of birth.

Alice Heim 4 test of General Intelligence (AH4)
Part I of the AH4 3 was administered to the oldest cohort of the Twenty-07 study at baseline. The AH4 was designed for use with a cross section of the adult population and part I consists of 65 questions which measure verbal and numeric cognitive abilities. It was administered according to the instructions in the test manual. The result used here is the total number of questions answered correctly within the time limit of 10 minutes. For comparability this total score is standardised to zero mean and unit standard deviation (zscored).
Vital status was ascertained via linkage to the UK's NHS central register. Participants of the Twenty-07 study were 'flagged' at the NHS central register at the outset of the study and since then regular notifications of deaths, embarkations and re-entries have been provided to the study. These notifications include date and cause of death. Primary and secondary causes are given in textual and coded forms, the latter being a mixture of ICD versions 9 and 10. The primary outcome here is the underlying cause of death grouped as shown in Supplementary Table 1. Of the analytic sample there were 820 deceased (men: 423) and 542 living by the census date (2 nd February 2017).

Covariates
Smoking status was classified as current smoker/ex-smoker/never smoked. The Social Class measure is derived from the occupation of the head of household coded to the six fold Registrar General's classification: Professional/Managerial and technical/Skilled non-manual/Skilled manual/Semiskilled/Unskilled. 4 Self-rated health (over the last twelve months) was recorded as excellent/good/fair/poor─this was dichotomised into excellent or good vs fair or poor. All covariates were treated as categorical when adjusted for.

Figure A
Hazard ratios (95% confidence intervals) for the association between a one SD higher score in intelligence test score at age 11 years and underlying cause of death in the Scottish Mental Survey 1947, N=65 765.
Hazard ratios (HR) are adjusted for age at intelligence testing, and sex.

Figure B
Hazard ratios (95% confidence intervals) for the association between a one SD higher score in intelligence test score at age 11 years and underlying cause of death by cancer type to age 79 years in the Scottish Mental Survey 1947. Hazard ratios (HR) are adjusted for age at intelligence testing, and sex, with the exceptions of ovarian and breast cancer (women only), and, prostate cancer (men only).  Hazard Ratios (and 95% confidence intervals) for a one SD higher score in intelligence score at age 11 years in relation to major causes of death in men and women separately in the Scottish Mental Survey 1947 Cause of death

Figure C
Hazard ratios (95% confidence intervals) for the association between a one SD higher score in intelligence test score at age 11 years and cause of death in the Scottish Mental Survey 1947, adjusting for school (N=65 765). Hazard ratios (HR) are adjusted for age at intelligence testing, and sex.

Figure D
Hazard ratios (95% confidence intervals) for the association between a one SD higher score in intelligence test score at age 11 years and cause of death by cancer type in the Scottish Mental Survey 1947, adjusting for school. Hazard ratios (HR) are also adjusted for age at intelligence testing, and sex, with the exceptions of ovarian and breast cancer (women only), and, prostate cancer (men only).

Figure E
Hazard ratios (95% confidence intervals) for the association between a one SD higher score in premorbid intelligence test score and underlying cause of death by older adulthood, in the Scottish Mental Survey 1947 and West of Scotland Twenty-07 cohorts. Hazard ratios (HR) are adjusted for age at intelligence testing, and sex.