Repeated intentional self-harm (SH) is associated with economic costs and increased risk of suicide. Estimates of repetition vary according to method of data capture and are limited by short periods of follow-up observation. Some sources use hospital records and others self-reported SH (SRSH). Our aim was to examine the relationship between SRSH and hospital-verified SH (HVSH) and later repetition of SH (predictive validity). We also aimed to examine whether rates of SH repetition differ between first-time presenters and non-first-time presenters using either definition of SH.

We conducted a large prospective study tracking SH attempts through an Accident and Emergency (A&E) department within the United Kingdom. We took a representative sample of 774 patients (30% of total who reported SH) and followed them for 5.6 years on average. The index episode of SH was recorded at the time of referral to staff in A&E. Prior episodes of SH were determined from an electronic search of A&E patient database, and in addition, recollection of prior SH as reported by the patient to their clinician at the time of index presentation was recorded.

Across the whole sample 32.0% of patients repeated SH within 1 year, which rose to 54.1% at completion of follow-up. Repetition rates were considerably higher in patients with a prior SH history than those presenting with a first SH episode after 1 year (47.9% vs. 19.6%) and by the end of follow-up (73.8% vs. 39.4%) (P<.001). Of 411 with self-reported first presentations, 45.2% repeated over the study period. In terms of predictive validity, 65.2% of those with previous SRSH repeated vs. 73.8% with previous HVSH (P<.001). There was low agreement between SRSH and HVSH (Kappa=0.353, 95% confidence interval 0.287–0.419, low).

We found relatively poor agreement between hospital-defined and self-reported SH. A total of 62.8% of those who denied SH actually had a hospital-verified previous episode. Patients with recorded prior SH and those who recall previous SH have significantly higher rates of repetition, but the two samples imprecisely overlap and predictive validity is stronger for HVSH.