Latent class analysis and ASA physical status yield differing results
Although perioperative risk stratification is often used to identify so-called independent risk factors, previous research has found that few are truly independent, and various combinations of comorbidities may have differential effects on mortality risk. Researchers turned to “latent class analysis,” a model-based clustering technique, and identified distinct classes of patients undergoing intraabdominal surgery that were significantly associated with risk for 30-day mortality.
“Latent class analysis is a statistical technique that has been occasionally used in other medical studies, such as in patients with dementia,” said Guohua Li, PhD, MD, professor of anesthesiology at Columbia University College of Physicians and Surgeons, in New York City. “But it has not been used in anesthesiology. So we thought it might be beneficial to use it in conjunction with some of our data to help determine risk.”
To that end, the investigators used the American College of Surgeons National Surgical Quality Improvement Program (ACS-NSQIP) database for 2005 to 2010 to obtain a cohort of patients undergoing intraabdominal general surgery. Demographic and comorbidity data were entered into statistical models to identify latent classes; statistical and clinical criteria were used to determine the optimal number of classes. Individuals were assigned to a class based on the highest posterior probability of class membership. Relative risk regression was used to find any associations between the latent classes and 30-day mortality, with adjustments for procedure and age.
“The goal of this exercise is to see if we could improve upon traditional risk measures by use of this technique, which incorporates the relationships between all comorbidities, instead of each one separately,” said principal investigator Minjae Kim, MD, assistant professor of anesthesiology, also at Columbia University College of Physicians and Surgeons.
As Dr. Kim reported at the 2015 annual meeting of the International Anesthesia Research Society (abstract S-318), the final data sample included 466,177 observations. A nine-class model was determined to be optimal for analysis. Based on prevailing demographic and comorbidity characteristics for each class, a general description of each class was obtained:
- Class 1: Age under 50 years, low rate of comorbidities and emergent procedures
- Class 2: Age under 50, morbidly obese females
- Class 3: Age under 60, low rate of comorbidities, obesity, no preoperative labs
- Class 4: Age under 50, normal to underweight females, emergent procedures
- Class 5: Age 40 to 70, low rate of comorbidities, with cancer
- Class 6: Age 50 to 70, morbidly obese with cardiovascular and associated comorbidities
- Class 7: Age over 50, mild to moderate comorbidities, emergent procedures
- Class 8: Age over 60, moderate to severe comorbidities, with cancer
- Class 9: Age over 60, severe comorbidity burden, functionally dependent
For the entire cohort, 30-day mortality was 2.6%, ranging from 0.06% to 21.9%. The highest-risk class (Class 9) had an eightfold greater rate of mortality than the group average (Table 1). Conversely, the lowest-risk class (Class 1) had a 41-fold lower mortality.
After adjustment for procedure and age, the latent classes remained significantly associated with 30-day mortality, although the rank order of some classes changed and others converged toward a similar adjusted risk (Table 2). At this point, the highest-risk class had a fivefold increase in mortality over the average-risk class, whereas the lowest-risk class had a 13-fold decrease in mortality.
As Dr. Kim said this type of analysis may ultimately prove most useful in stratifying patients in conjunction with American Society of Anesthesiologists (ASA) physical status. “When we’re using ASA class, it’s really the clinician’s gut feeling,” he said, “and my gut feeling can be different from someone else’s gut feeling, so there are some interrater reliability issues. Our gut feeling is generally very good, but we can improve upon this gut feeling with an objective analysis of comorbidity information.
“For example, in this analysis, ASA 2 patients had an overall mortality risk of 2.3%,” he noted. “But after separating the latent classes, we identified a mortality risk range for ASA 2 patients between 0.4% and 11%.”
Given the ability of latent class analysis to include preoperative data to stratify patients by risk for perioperative mortality, the researchers saw it as a potential tool to be incorporated into all anesthesiologists’ practice, despite its apparent complexity. Indeed, Drs. Li and Kim foresaw a time when a computer program would determine patient mortality risk based on preoperative assessment data. “We could devise a simple algorithm in the electronic data record that allows for computer systems to automatically generate risk classes based on data,” Dr. Li noted.
Making the tool an automated part of the clinician’s assessment is an important part in its ultimate success, Dr. Kim added. “Comorbidity indices are very easy to calculate at the bedside, but I don’t see a lot of clinicians actually using them. So going forward, I think the best use would be to have it embedded in the medical record system so that when you click on the patient’s record, it just automatically incorporates all the available data and creates a scale.”