OBJECTIVES: To describe the pre-diagnostic pathway of patients with non-tuberculous mycobacterial lung disease (NTMLD) in primary care in the UK and to use this information to develop a predictive model to identify individuals likely to have NTMLD.

METHODS: Patients were selected from The Health Improvement Network (THIN), a large UK primary-care database. Incident patients with NTMLD between 2003 and 2017 with THIN records were identified through: (i) diagnosis in primary care, (ii) diagnosis in secondary care and (iii) having received an NTMLD treatment regimen for at least 180 days. All patients had at least three years of medical history. Metrics describing the frequency and timing of the events were passed to a gradient boosting trees predictive model. Using the model, we estimated the prevalence of undiagnosed individuals likely to have NTMLD.

RESULTS: Annual prevalence of NTMLD increased steadily in the period from 2006 to 2016 from 2.7 to 4.8 per 100,000. The selected patient population comprised 741 NTLDM patients and a random stratified sample of 112,874 of the non-NTMLD patient population. The average age was 60 years and 48 years for NTMLD and non-NTMLD patients, respectively. The most common diagnoses observed in the NTMLD cohort were chronic obstructive pulmonary disease (34%), asthma (25%), tuberculosis (28%) and bronchiectasis (13%).Using the model to identify individuals likely to have NTMLD and considering risk thresholds of 0.90 to 0.95, we estimated that the total prevalence of diagnosed cases and individuals likely to have NTMLD in 2016 to range from 9 to 16 in 100,000.

CONCLUSIONS: The data that are captured in THIN enable a predictive model to be developed to identify potentially undiagnosed cases of NTMLD in primary care in the UK.