The prevalence of frailty in a community-dwelling population has been estimated to be between 6.9% and 10.7% in large studies and meta-analyses, and is substantially higher in cohorts of thoracic surgical patients.27–29 Although frailty is commonly recognized by surgeons as a risk factor for poor patient outcomes, challenges in measuring frailty in the preoperative setting remain a barrier to its use, such as the need for specialized tests and the difficulty in choosing among the many options. In many cases, surgeons continue to rely on the “eyeball test” to subjectively assess a patient’s fitness for surgery, which has been shown to be inferior to statistical risk estimates such as the VA risk estimate for cardiac surgery and the American College of Surgeons NSQIP Surgical Risk Calculator, which is widely used to estimate risk of death and complications after general surgery operations.30
The use of a frailty index based on a deficit accumulation assessment was advanced by Mitnitski et al. in 20017 and has since been adapted to both general surgical and thoracic surgery populations. Recently, there has been an interest in utilizing EHR data to develop frailty scores using routinely collected data to minimize clinician and patient burden.31,32 However, the operationalization of a frailty index based on the EHR has gained limited traction due to the multitude of scoring systems, reliance on clinical terminology codes, and the limited availability of specialized serum biomarkers in patients undergoing operations.
The frailty indices that we selected to investigate have been previously validated in large patient cohorts, though the patient population in each study was somewhat different. The mFI-5 used data from the NSQIP 2015 database in patients undergoing surgery22, while the FI-Laboratory23 was validated in hospitalized medical patients alone, and the FI-Lab26 and PNI33 were validated in hospitalized and community-dwelling patients. Increases in each of the four indices were associated with increased mortality in each study. The variable nature of patient settings in these studies was likely the driving factor behind the availability of data for the present study. The mFI-5 was able to be calculated in all patients in our study because comorbidities and ASA class are frequently collected for surgical patients. On the other hand, a minority of patients had sufficient data for the FI-Lab, likely stemming from the fact that a majority of patients in that index study were community-dwelling and were participants in a cohort study in which a specific set of labs were prospectively determined and collected.
Frailty is a risk factor for increased hospital length of stay, hospital complications, readmission, and death.34,35 Thus, we attempted to use comorbidity, vital signs, and lab data routinely collected preoperatively to test the applicability of existing frailty scores in a cohort of thoracic surgery patients. We found that only the mFI-5 could be calculated within 30-days of surgery, and that the FI-Laboratory, PNI and FI-Lab could be calculated in only a minority of patients within six months of surgery using routinely available EHR data. Although these scoring systems were previously validated in a non-lung resection patient population, we found that most had limited utility in predicting postoperative complications and death in a cohort of thoracic surgery patients undergoing lung resection at a single, urban academic medical center.
The mFI-5 was the most robust index in the present study, both in terms of the ability to calculate scores for patients and its utility in predicting complications, specifically pulmonary and overall complications. The further benefit of the mFI-5 is its ease of use, especially in a surgical population in which comorbidities are elicited and ASA class is assigned preoperatively. As such, the mFI-5 could be considered in the preoperative evaluation of patients undergoing major thoracic or abdominal surgery, although further investigation is necessary into whether EHR-based mFI-5 scores can be generalized to surgical populations as a whole.
Our study has a number of potential limitations, including its retrospective nature, relatively small sample size, and the long time frame over which patients were included. We analyzed preoperative and perioperative complication data for patients as early as 1995, and changes in surgical technique and postoperative care over that time period may have led to changes in the frequency and type of complications, limiting significant findings.
In conclusion, the availability of extant data in the EHR does not permit calculation of many frailty scores possibly related to surgical risk. A possible exception is the mFI-5 score, which we found to be predictive of some outcomes after major lung resection in our population. Further evaluation of mFI-5 in larger retrospective sample or prospective cohorts may be useful. Identifying a reliable EHR-based frailty assessment tool could be helpful determining which patients are a higher preoperative risk for adverse events. Using this data, surgeons may be able to preemptively mitigate a patient’s surgical risk with appropriate perioperative resources, referrals, and prehabilitation.