Algorithms Using Administrative Data Found to Be Accurate at Detecting AECOPD Events

Validity assessments revealed that 2 algorithms utilizing administrative claims data and electronic medical records were accurate at identifying moderate and severe events signaling acute exacerbation of chronic obstructive pulmonary disease (AECOPD).

A study evaluating 2 utilization-based algorithms designed to detect moderate and severe acute exacerbations of chronic obstructive pulmonary disease (AECOPD) showed sufficient validity, suggesting that they could prove useful in health outcomes research and quality improvement projects.

The study, which was published in the International Journal of Chronic Obstructive Pulmonary Disease, adds to the current evidence by demonstrating the validity of using large administrative databases and electronic health records (EHRs) containing health care utilization data and further outlines objective criteria that providers can use to identify and confirm AECOPD events.

“This is an important issue, as accurate identification of AECOPD events in electronic administrative data is essential for improving population health surveillance and practice management,” wrote the investigators.

Exacerbations are significantly linked to increases in mortality risks and health care resource utilization in the form of hospitalizations, visits to emergency departments (EDs) or urgent care centers, and prescription medications. Patients who experience more frequent exacerbations could experience acute irreversible loss of lung function and a reduced quality of life. Thus, preventing exacerbations is the primary goal when treating COPD.

Although administrative data can provide valuable health care utilization information, there is no consensus on how to best use administrative data for identifying exacerbations, which represents a significant research barrier for retrospective analyses of large health care databases.

The retrospective longitudinal cohort study took place between January 1, 2010, and September 30, 2015. A baseline period of 12 months prior to the index date was established, and observation of patients spanned from the index date until the patient died, unenrolled from the study, or the study period ended.

To be included, patients had to have at least 1 hospitalization or ED visit or at least 2 outpatient visits during the baseline period. They also had to be receiving at least 1 prescription for maintenance therapy, have continuous health insurance coverage before and after the index date, and have available EHR data.

Health insurance claims and patient data were collected from 2 independent health systems: Kaiser Permanente Mid-Atlantic States (KPMAS) and Reliant Medical Group. In total, 7914 patients 40 years or older with COPD were enrolled (KPMAS, n = 2366; Reliant, n = 5548). The patients in the KPMAS system had a mean (SD) age of 72.7 (11.5) years and 56.6% were women whereas those in the Reliant system had a mean (SD) age of 68.9 (10.7) years and 54.7% were women.

The investigators developed 2 algorithms for examining administrative data: 1 that would identify moderate AECOPD events and 1 that would identify severe AECOPD events. The algorithms were validated by calculating the positive predictive values (PPVs) and the negative predictive values (NPVs), which represent the probability of a positive or negative result being accurate, respectively.

Among the moderate AECOPD events, 98.3% (293 of 298) of the events detected by the algorithm were true positives. Similarly, 96.0% (216 of 225) of the severe AECOPD events identified by the algorithm were true positives.

The PPV for severe AECOPD events was lowest for the patients with a diagnosis of chronic airway obstruction, not elsewhere classified (94.0%; 95% CI, 86.7-98.0). When evaluating accuracy by COPD status (verified, unverified, or undefined), the PPV was greater than 93.8% across all categories for both algorithms.

The NPV was lower for the algorithm for moderate AECOPD events compared with the one for severe AECOPD events (75.0%; 95% CI, 65.3-83.1; vs 95.0%; 95% CI, 88.7-98.4). When comparing health systems, the NPV for moderate exacerbations was 72.0% for Reliant (95% CI, 57.5-83.8) and 78.0% for KPMAS (95% CI, 64.0-88.5). The NPV values for severe exacerbations was 90.0% for Reliant (95% CI, 78.2-96.7) and 100% for KPMAS (95% CI, 92.9-100.0).

To explain the lower NPV values for moderate exacerbations, the investigators said that missing claims data, timing, and the potential for the algorithm to not account for secondary diagnoses as indications of AECOPD possibly contributed. Another potential cause may have been that some databases did not account for programs that provide patients with prefilled medications for them to take at the onset of symptoms and indicate patients to update a provider when they administer the medication.

“Services like these are becoming more prevalent, meaning that claims-based algorithms that do not include them have the potential to miss moderate AECOPD events,” wrote the investigators.

Several limitations were identified, including that EHR data are not collected for research purposes, the results may not be generalizable to other systems or administrative databases, the algorithms may perform differently in other systems, and that there was a small potential for misclassification of patients into the COPD cohort.


Mapel DW, Roberts MH, Sama S, et al. Development and validation of a healthcare utilization-based algorithm to identify acute exacerbations of chronic obstructive pulmonary disease. Int J Chron Obstruct Pulmon Dis. 2021;6:1687-1698. doi:10.2147/COPD.S302241