For acute/potentially acute conditions (myocardial infarction, stroke, congestive heart failure, hemiplegia, falls, weight loss/undernutrition) which could have occurred as a result of the CAP, records from any point prior to but excluding the CAP incident date itself were used as evidence of a pre-existing condition.
The primary exposure of interest was NSAID use in the 5 years before the index date, excluding the year before the index date.
The median start date was in 2006 (range 1997 2011); when studies evaluating discussion forums were excluded, the start date was more recent (median 2008, range 2000 2011).
The event of ADs was defined as three times of outpatient visits or one inpatient admission with an AD diagnosis (ICD-9-CM code: 300.xx were included but 300.4 dysthymic disorder was excluded) between the date of TBI diagnosis and 31 December 2010.
For each member of the cohort we also extracted all of their linked hospital admissions for any reason (public and private) in the 20 years prior to the index date, but excluding same-day separations for routine dialysis (ICD9-Clinically Modified code V56, ICD10-AM code Z49) [ 13- 15].
The ICD-IX codes of interest were reported in the death certificates of 599 subjects (one subject was subsequently excluded because the date of birth was unknown).
However, interruption events were excluded if the date was missing.
Sixty-seven (29%) were excluded because the date of infection could not be estimated with sufficient precision.
These latter patients were excluded as the date of starting metformin was unknown (no prescribing data available prior to 1993).
Three cases were excluded as the date of hospital admission was more than four months after the date of onset of paralysis.
