The effect of using NHS number as the unique identifier for patients who self-harm: a multi-centre descriptive study

BACKGROUND Processing personal data for research purposes and the requirement of anonymity has been the subject of recent debate. We aimed to determine the proportion of individuals who present to emergency departments with non-fatal suicidal behavior where an NHS number has been successfully traced and to investigate the characteristics of patients associated with non-capture. METHOD This was a descriptive study of people attending after self-harm using allocation of NHS numbers as main outcome measurement. Data from the Multicentre Monitoring of Self-Harm Project from 3 centres in England were used to identify consecutive patients (N = 3000) who were treated in six emergency departments in Oxford, Manchester and Leeds in 2004 and 2005 following self-harm. RESULTS NHS number was available between 55-73% of individuals across centres. Characteristics associated with non-recording of NHS number in more than one centre included those from ethnic minority groups (Oxford: chi-squared statistic = 13.6, df = 3, p = 0.004; Manchester: chi-squared statistic = 13.6, df = 3, p </=0.001) and the homeless or living in a hostel or other institution (Oxford: chi-squared statistic = 40.9, df = 7, p = <0.001; Manchester: chi-squared statistic = 23.5, df = 7, p = 0.001). Individual centre characteristics included being of male gender (Leeds: chi-squared statistic = 4.1, df = 1, p = 0.4), those under 25 years (Oxford: chi-squared statistic = 10.6, df = 2, p = 0.005), not being admitted to general hospital (Leeds: chi-squared statistic = 223.6, df = 1, p </=0.001) and using self-injury as a method of harm (Leeds: chi-squared statistic = 41.5, df = 2, p </=0.001). CONCLUSION Basing research studies on NHS number as the unique identifier, as suggested by the Data Protection Act 1998 and the Patient Information Advisory Group, would exclude some of the most vulnerable groups for further self-harm or suicide. This bias may also affect other research registers.