Predicting Mortality in Community Acquired Pneumonia – Dr. Robert Centor Interviews PSI Creator Dr. Michael Fine

Centaur Centor3 (1)

Centor’s Corner

Dr. Michael Fine, professor of medicine at the University of Pittsburgh, led the team that developed the Pneumonia Severity Index (PSI) and began studying the prognosis and other clinical aspects of community-acquired pneumonia (CAP) in the early 1990s.

His interest in predicting mortality in CAP started while he served as chief resident in internal medicine at the University of Pittsburgh. His mentor, Dr. Wishwa Kapoor, then hired him after his general internal medicine fellowship in the Harvard Generalist Faculty Development Program.  At the time Dr. Fine transitioned from fellowship to faculty at the University of Pittsburgh, the Agency for Health Care Policy and Research (now the Agency for Healthcare Research and Quality, AHRQ) had a well-funded portfolio of research projects called PORT (Patient Outcome Research Teams) studies.  

Dr. Fine

At that time, I served on the study section that reviewed the Pneumonia PORT proposal that Drs. Fine and Kapoor submitted.  After the study received funding, I followed their work closely.  

The Pneumonia PORT team developed the PSI in a series of retrospective and prospective cohort studies. The PSI predicts 30-day all-cause mortality using age, gender, concurrent diseases, mental status, vital signs and laboratory values.  While comprised of 20 individual prognostic variables, in 2017 we have great calculators that make using the PSI quick and accurate.  

After deriving the PSI in a national retrospective cohort of over 14,000 patients with CAP, Dr. Fine and colleagues validated the accuracy of risk stratification based on this prediction rule for prognosis in two independent patient cohorts. They applied the PSI to a cohort of 38,000 patients hospitalized in 193 hospitals in the State of Pennsylvania and 2,287 patients enrolled in the Pneumonia PORT prospective cohort study.  Since development of the PSI, multiple studies from North America, Europe, and Asia have validated this prediction model.


Dr. Centor

The prediction model assigns the patient to one of five risk groups.  The two lowest risk groups have death rates less than 1%.  Dr. Fine recommends considering outpatient treatment in these patients, but cautions that physician judgment should always be used in conjunction with the PSI risk class.

Risk class 3 is the most challenging.  In the original study, these patients had a mortality risk of 0.9 to 2.6%. For this risk class in particular, physicians must use good clinical judgment and consider the patient’s living situation and level of social support.  Many of these patients will do well with outpatient treatment, yet some may need to be hospitalized for a variety of reasons.

Risk class 4 patients have observed mortality rates of 8.2 to 9.3%, while class 5 patients have rates of 27% or higher. These patients clearly need hospitalization, and many require higher intensity levels of inpatients care, including intensive care unit admission.

I asked Dr. Fine to define CAP.  He said that we should look for classic signs and symptoms of a lower respiratory tract infection and the presence of a new chest X-ray infiltrate or air space disease. We then discussed the differential diagnosis of CAP.  When the patient has already received treatment for CAP, and still has signs, symptoms, and an abnormal chest x-ray, we should consider other infectious and non-infectious diagnoses.

Many physicians prefer to use the CURB-65 Score rather than the PSI to assess pneumonia severity, primarily because they can remember the CURB-65 and calculate it without taking their smartphone out of their pocket.  However, Dr. Fine gives two strong reasons for using the PSI rather than the CURB-65 for this purpose:

First, the PSI has better discriminatory power.  Statisticians use the ROC (receiver operating characteristic) curve area (the same number as the C-statistic) to compare discrimination.  We can understand the ROC area (or C-statistic) as the result of the 2-pair forced choice problem.  If one randomly takes one patient who lived and one patient who died within 30 days, the probability that the score would predict which patient lived is the ROC area.  Thus, these numbers range from 0.5 to 1.  A perfect test has an ROC area of 1; a coin flip gives an ROC area of 0.5.  In comparison studies, the PSI consistently has a higher ROC area.

Additionally, a series of randomized controlled trials supported by numerous quasi-experimental studies (pre-post interventions) have shown that using the PSI to guide the initial site of treatment (home versus hospital) safely decreases hospitalization of low-risk patients.  Dr. Fine knows of no such studies using the CURB-65.

The argument against using the PSI, its complexity, was much more important prior to smartphones and medical calculators. He strongly suggests that after diagnosing CAP, physicians should estimate risk using the PSI.  As a clinician, he cautions that clinical judgment should always be used to supplement the calculator. Cases in which clinical judgment and the PSI suggest different treatment strategies should always rely on patient safety in driving the treatment decision.