ľֱ

Could Data Find PAH Before Doctors Do?

— Algorithm based on insurance claims accurate in flagging condition 6 months prior to diagnosis

MedpageToday

SAN FRANCISCO -- Patients' utilization data suggested a new approach to identifying those who may have pulmonary arterial hypertension (PAH) earlier than is possible with current methods, a researcher said here.

A machine-learning algorithm drawing on individuals' insurance claims was 73% accurate in identifying patients with PAH 6 months before their doctors were able to make formal diagnoses, said Katherine Bettencourt, PhD, a data scientist at Johnson & Johnson's Janssen subsidiary in Titusville, New Jersey.

Based on data from 1,724 people diagnosed with PAH and 5,352 otherwise similar individuals with non-PAH respiratory or cardiovascular diagnoses, the algorithm highlighted a number of factors linked to PAH -- and present 6 months before they were formally diagnosed -- that are not part of the conventional diagnostic checklist, Bettencourt reported at the American Thoracic Society's annual meeting.

These factors included:

  • Time from first symptom
  • Percent of total claims related to circulatory issues
  • Number of unique circulatory codes
  • Number of prescriptions written
  • Circulatory diagnoses in patient's history

PAH patients also tended to have greater utilization in terms of total cost, more imaging procedures, and more comorbidities than the non-PAH controls -- again, as seen 6 months prior to the ultimate diagnosis.

The most common circulatory codes appearing in PAH patients' records, and more common than in controls, were essential hypertension, heart failure, coronary artery disease, and what Bettencourt's group summarized as "complications and ill-defined descriptions of heart disease." These were seen in 18%-54% of those determined to have PAH.

PAH patients included in the study appeared generally representative of those in the larger U.S. population. All regions were represented, about two-thirds were women, and a similar proportion were white. Mean age was about 70. As expected with this age distribution, some 70% of PAH patients were on Medicare; the rest had commercial insurance.

After training, the algorithm's area under the receiver-operating characteristic curve was 0.84, and the "recall rate" -- the fraction of "true" PAH patients identified as such by the algorithm -- was 0.73.

While these figures are not good enough for the algorithm to serve as a diagnostic in and of itself, its performance "indicates the feasibility of identifying patients at a population level who might benefit from PAH-specific screening," using "routinely collected" claims data, Bettencourt indicated.

That would allow for earlier diagnosis than is now generally the case, she said.

One particular and important benefit could be reducing hospitalizations: Bettencourt noted that the claims data showed a dramatic uptick in hospitalizations during the 6 months prior to patients' actual PAH diagnosis. During months 6-12 before diagnosis, 20% of the PAH patients were admitted (vs 11% of controls), which jumped to 42% over the next 6 months. Many of those hospitalizations could have been prevented if PAH was already recognized and under treatment, she suggested.

Bettencourt also picked out rates of imaging procedures as an important signal, even though it made only a modest contribution to the algorithm. The mean rate was 6.5 per PAH patient, as compared with 3.6 among controls.

This "highlights the need for improved screening and diagnostic processes," she said.

That was also one of the lessons from another , by researchers from numerous institutions in the U.S. and Great Britain, showing that mortality from PAH has skyrocketed since 1999. Age-adjusted death rates fully doubled over the following two decades for men and for women (and was always about 50% greater in women), for white and for Black Americans, and in every major U.S. region.

Black women had the highest rates, at 4.0 per 100,00 population in 2019, the study's final year.

The study's authors concluded that these data support "an urgent need" for more population-level research into PAH and ways to reduce its burden.

  • author['full_name']

    John Gever was Managing Editor from 2014 to 2021; he is now a regular contributor.

Disclosures

The study was funded by Janssen and all authors were employees there or in its Actelion unit.

Primary Source

American Thoracic Society

Bettencourt K, et al "A claims-based, machine-learning algorithm to identify patients with pulmonary artery hypertension (PAH)" ATS 2022; Poster 617.