BACKGROUND: UK primary care records are computerised and these records are used for both research and quality improvement. However, there is disparity in the prevalence of diabetes found in epidemiological studies compared with that reported through the UK's national quality improvement scheme. OBJECTIVE: To investigate how non-diagnostic computer data could be used to identify, confirm or refute prevalent cases of people with diabetes. METHOD: We carried out a literature review to identify the most accurate non-diagnostic markers. For each type of diabetes we focused on four broad areas; demographic details, biochemical markers, clinical features and therapeutic strategies. Sample markers were tested by calculating their positive predictive value (PPV) and sensitivity (Sn) and their ability to differentiate between types of diabetes. RESULTS: Biochemical markers were useful in identifying cases of diabetes but not in differentiating between types of diabetes as the same plasma glucose criterion is used to diagnose Type 1, Type 2, and 'other' types of diabetes; the lack of a 'fasting' qualifier blunts the use of this marker. Auto-immune markers were the most accurate in identifying Type 1 diabetes but are not recorded frequently in primary care. Clinical features of diabetes and therapeutic strategies were of some use--however, without time sequence data are difficult to interpret. Raised plasma glucose (PG), and glycated haemoglobin (HbA1c), had useful PPV but low Sn. When PG was more than 7.0 and less than 11.1 mmol/l, PPV equalled 77.8% and Sn 48%; and when PG was 11.1, PPV equalled 92% and Sn 17%. For an HbA1c of more than 6.5%, PPV was 89% and Sn 73.3%, and for an HbA1c of more than 8, PPV was 92% and Sn 26%. A person with a combination of aged under 30 years and prescribed insulin has an 84% PPV of Type 1 diabetes; if they also have a BMI <30 kg/m2 the PPV increases to 88%. A person age over 45 years and with a BMI >30 kg/m2 has a 5.3% PPV of Type 2 diabetes; if they are also hypertensive the PPV is 30%; Asian ethnicity increases PPV to 44%. CONCLUSION: Non-diagnostic data has the potential to confirm or refute the diagnosis of diabetes and identify its type.
BACKGROUND: UK primary care records are computerised and these records are used for both research and quality improvement. However, there is disparity in the prevalence of diabetes found in epidemiological studies compared with that reported through the UK's national quality improvement scheme. OBJECTIVE: To investigate how non-diagnostic computer data could be used to identify, confirm or refute prevalent cases of people with diabetes. METHOD: We carried out a literature review to identify the most accurate non-diagnostic markers. For each type of diabetes we focused on four broad areas; demographic details, biochemical markers, clinical features and therapeutic strategies. Sample markers were tested by calculating their positive predictive value (PPV) and sensitivity (Sn) and their ability to differentiate between types of diabetes. RESULTS: Biochemical markers were useful in identifying cases of diabetes but not in differentiating between types of diabetes as the same plasma glucose criterion is used to diagnose Type 1, Type 2, and 'other' types of diabetes; the lack of a 'fasting' qualifier blunts the use of this marker. Auto-immune markers were the most accurate in identifying Type 1 diabetes but are not recorded frequently in primary care. Clinical features of diabetes and therapeutic strategies were of some use--however, without time sequence data are difficult to interpret. Raised plasma glucose (PG), and glycated haemoglobin (HbA1c), had useful PPV but low Sn. When PG was more than 7.0 and less than 11.1 mmol/l, PPV equalled 77.8% and Sn 48%; and when PG was 11.1, PPV equalled 92% and Sn 17%. For an HbA1c of more than 6.5%, PPV was 89% and Sn 73.3%, and for an HbA1c of more than 8, PPV was 92% and Sn 26%. A person with a combination of aged under 30 years and prescribed insulin has an 84% PPV of Type 1 diabetes; if they also have a BMI <30 kg/m2 the PPV increases to 88%. A person age over 45 years and with a BMI >30 kg/m2 has a 5.3% PPV of Type 2 diabetes; if they are also hypertensive the PPV is 30%; Asian ethnicity increases PPV to 44%. CONCLUSION: Non-diagnostic data has the potential to confirm or refute the diagnosis of diabetes and identify its type.