Developing an algorithm to identify people with Chronic Obstructive Pulmonary Disease (COPD) using administrative data

Abstract

Background: An important prerequisite for the Chronic Care Model is to be able to identify, in a valid, simple and inexpensive way, the population with a chronic condition that needs proactive and planned care. We investigated if a set of administrative data could be used to identify patients with Chronic Obstructive Pulmonary Disease in a Danish population.

Methods: Seven general practices were asked to identify patients with known Chronic Obstructive Pulmonary Disease in their practices. For the 266 patients (population A), we used administrative data on hospital admissions for lung-related diagnoses, redeemed prescriptions for lung-diseases drugs and lung- function tests combined to develop an algorithm that identified the highest proportion of patients with Chronic Obstructive Pulmonary Disease with the fewest criteria involved. We tested nine different algorithms combining two to four criteria. The simplest algorithm with highest positive predictive value identified 532 patients (population B); with possible diagnosis of Chronic Obstructive Pulmonary Disease in five general practices. The doctors were asked to confirm the diagnosis. The same algorithm identified 2,895 patients whom were asked to confirm their diagnosis (population C).

Results: In population A the chosen algorithm had a positive predictive value of 72.2 % and three criteria: a) discharged patients with a chronic lung-disease diagnosis at least once during the preceding 5 years; or b) redeemed prescription of lung-medication at least twice during the preceding 12 months; or c) at least two spirometries performed at different dates during the preceding 12 months. In population B the positive predictive value was 65.0 % [60.8;69.1 %] and the sensitivity 44.8 % [41.3;48.4 %)] when the "uncertain" were added to where doctors agreed with the diagnosis. For the 1,984 respondents in population C, the positive predictive value was 72.9 % [70.8;74.8 %] and the sensitivity 29.7 % [28.4;31.0 %].

Conclusions: An algorithm based on administrative data has been developed and validated with sufficient positive predictive value to be used as a tool for identifying patients with Chronic Obstructive Pulmonary Disease. Some of the identified patients had other chronic lung-diseases (asthma). The algorithm should mostly be regarded as a tool for identifying chronic lung-disease and further development of the algorithm is needed.

Trial registration: www.clinicaltrials.gov (NCT01228708).

Figures

Figure 1

Patients identified by algorithm and verified by GPs - population B. Flowchart showing the number of patients for which GPs can verify the diagnosis of COPD when the patients have been identified by the COPD algorithm developed in the study.

Figure 2

Positive Predictive Values when patients verified by GPs in population B. The Positive Predictive Value when GPs verified the COPD diagnosis for patients identified by the algorithm developed in the study. The patients were divided into ten-year age groups.

Figure 3

Patients who verified their COPD diagnosis - population C. Flowchart showing the number of patients who verified their COPD diagnosis when identified by the algorithm developed in the study.

Figure 4

The Positive Predictive Value when patients verified their COPD diagnosis in population C. The patients were identified by the algorithm developed in the study and divided into ten-year age groups.