More than 300 studies have been performed to assess the positive predictive value of the coding of diseases for the presence of the coded diagnosis. These have in general found high positive predictive values in the order of 90% [7, 8]. It is likely that the same will be true for Autoimmune Hepatitis, but that is not what this study has examined. We in contrast have examined the sensitivity of the coded diagnosis of a chronic disease for its presence, and this to our knowledge has very rarely been done before [9].
We assessed the validity of the coding of AIH using a cohort of 42 patients with AIH and 7 patients with other diagnoses or an overlap syndrome whose biopsies were reviewed at the liver biopsy meeting at the Queens Medical Centre, Nottingham, UK, between 2004 and 2009. The diagnosis of AIH is complicated, as essentially there is no single pathognomonic feature [1] and the clinical context is of the utmost importance. Despite there being validated scoring criteria, the consensus of recent British Society of Gastroenterology guidelines on AIH is that this diagnosis requires considerable “clinical expertise” [1] and scoring should be used as an adjunct. We chose to use a consultant diagnosis of autoimmune hepatitis as our gold standard as had Yeoman [10]. We had 42 patients available to us who had a clinical diagnosis of AIH as confirmed by the managing hepatologist, for whom we requested coding information from GPs. We received 39 responses, 37 of the patients had coded or free text information available for the diagnosis of autoimmune hepatitis. Two additional practices wrote back to us to state the patient had left the practice or that they patient was deceased and they no longer held their electronic or paper records (Table 4). Of the 37 GP practices which had information available on the electronic recording of patient data, 89% used a specific READ code for Autoimmune Hepatitis (J638.00 Autoimmune Hepatitis). Others had used non-specific codes which may be interpreted as other forms of liver disease or used free text (Table 4).
We believe therefore that the sensitivity of the coded diagnosis of Autoimmune Hepatitis for the detection of this condition in General Practice is 89% i.e. 89% of all people diagnosed with Autoimmune Hepatitis in hospital get a specific code in electronic primary care records for Autoimmune Hepatitis. This high sensitivity ensures that electronic records are unlikely greatly to underreport the incidence and prevalence of AIH and thus such records are suitable for studying the epidemiology of AIH in databases such as the General Practice Research Database, QResearch or the Health Improvement Network in the United Kingdom. Mediplus is a general practice database used in France, Austria and Germany; Holland uses PHARMO, IPCI and PALGO. In addition there are hospital based databases in England, Spain, Sweden Canada, Australia, Asia (China, Hong Kong, Japan, South Korea, Taiwan and Singapore), Latin America (Brazil, Argentina, Mexico, Venezuela, Peru, Chile). Evidence exists that data from the UKGPRD correlates with other databases in Europe e.g. fracture risk in coeliac disease was found to be increased to the same extent in both the UK GPRD [11] and in the Swedish National Inpatient Register [12] There are also Danish Registries which have been used to look at liver diseases [13].
Seven of the initial fifty patients were subsequently found to have other diagnoses, or overlap syndromes. We wrote to their general practitioners to assess the possibility of a miscoded diagnosis of AIH. If we were considering performing population-based studies, we would be able to exclude most of the overlap and misdiagnosed patients by excluding any person with a code for another form of liver disease as well as autoimmune hepatitis. 71% of these overlap syndromes or those with subsequent diagnoses were accurately coded.
Limitations
A limitation of this study in the assessment of validity is that it assessed sensitivity of AIH alone. Other measures of diagnostic accuracy or utility include positive predictive value, specificity and negative predictive value. We believe the positive predictive value of AIH to be high in electronic primary care data because this has been shown to be the case in other gastrointestinal diseases such as Coeliac Disease, IBD, Cirrhosis [14] all of which require a similar diagnostic process to AIH including assessment in secondary care and a combination of clinical judgement, laboratory and histological results. Nonetheless further work looking at the positive predictive value of this diagnosis in the GPRD, validating against Hospital Episode Statistics records and/or the clinic letters/discharge information available to the General Practitioners would enhance our confidence in the value of these data. We believe the negative predictive value of a disease such as AIH which has an incidence of approximately 1.9/100 000 would be high simply because it is so uncommon. Other limitations include the lack of data on a therapeutic response, however, as this was a retrospective study there was insufficient information available to come to a valid conclusion about therapy.
All the patients were sourced via the liver biopsy meeting in Nottingham University Hospitals. In Nottingham it is unit policy to biopsy all patients in whom a diagnosis of AIH is made in view of the potentially lifelong therapy which is being embarked upon. Hence in Nottingham only patients who decline a biopsy will not be biopsied. The method chosen to obtain patients for the audit was recommended by the departmental hepatologists as it would be least likely to miss any cases. When compared to a non-transplant centre’s experience of autoimmune hepatitis in Sheffield, only 16 out our 245 cases did not have an initial biopsy, and eventually 7 underwent a biopsy [3].