Computerized pneumonia severity index identifies patients at low risk for death

Computerized versions of the Pneumonia Severity Index accurately identified patients with community-acquired pneumonia at low risk for death, according to data published in the Annals of the American Thoracic Society.

“Current advances in the electronic health record and in clinical prediction modeling may make consistent severity assessment more feasible and recent guidelines have called for the evaluation of computerized approaches,” Barbara E. Jones, MD, physician in the division of pulmonary and critical care at Veterans Affairs Salt Lake City Health Care System at the University of Utah, Salt Lake City, and colleagues wrote. “Computer-assisted risk assessments can reduce the burden and incorporate more complexity than those designed for manual calculation.”

Researchers evaluated 297,498 adults with community-acquired pneumonia (median age, 68 years; 95% men) who presented to EDs at 117 Veterans Affairs Medical Centers from January 2006 to December 2016. Researchers compared a computerized score of the Pneumonia Severity Index that contained all variables except confusion and pleural effusion score with 10 novel models that incorporated additional patient factors.

The primary outcome was all-cause mortality within 30 days of the initial ED visit. Secondary outcomes included hospitalization within 24 hours of the initial ED visit and 7-day secondary hospitalization within 7 days of discharge from the ED.

The median Pneumonia Severity Index score was 86. Sixty percent of patients were hospitalized. The 30-day all-cause mortality rate was 6.6%. In addition, 7% of 179,682 patients initially treated and discharged from the ED returned with secondary hospitalization within 7 days.

Seven percent of the 297,498 encounters resulted in death within 30 days of the initial ED visit.

Performance of the models increased with greater complexity of the models and increased number of variables. The area under the curve (AUC) for predicting 30-day mortality was 0.77 for the computerized Pneumonia Severity Index model. High performance was also observed among models limited to age, sex and physiologic variables.

With a Pneumonia Severity Index score threshold of 970 and a mortality risk cutoff of < 2.7%, the computerized Pneumonia Severity Index score identified 31% of all patients who had a lower risk for all-cause mortality, while the boosted decision-tree algorithm machine learning Pneumonia Severity Index model using the Extreme Gradient Boosting algorithm with age, six and 26 other physiologic factors identified 53% of all patients as lower risk and the same boosted machine learning model with all 69 variables identified 56% of all patients as low risk, according to the results. The models determined similar rates of mortality, hospitalization and 7-day secondary hospitalization.

“As our health care systems advance, so do the ways we conceptualize illness severity. The advantage of the original [pneumonia severity index] is that it is familiar, well-

validated heuristic that provides consistency that can reduce inequities and unwarranted variation despite increasing information and the demands placed on clinicians,” the researchers wrote. “The disadvantage is that it is both burdensome and oversimplified: clinicians take far more features into consideration when assessing illness severity and the amount of clinical data that can be accommodated — with the help of computers — has increased.”