Sarcopenia, Obesity, and Sarcopenic Obesity: Retrospective Audit of Electronic Health Record Documentation versus Automated CT Analysis in 17 646 Patients

Abstract

Introduction

Sarcopenia and obesity are potentially reversible conditions associated with numerous adverse clinical outcomes, including death (1). Although the definition of obesity is widely known, physician knowledge of sarcopenia is limited (2). Since inception of the term sarcopenia in 1989, the definition has consistently included reduced muscle mass and clinically it is associated with reduced muscle strength (3).

Similar to obesity, guidelines for sarcopenia are now established by expert groups for specific prescriptions of exercise and nutrition (4). Despite these guidelines, no prior study has reported an audit of the electronic health record (EHR) using CT as the reference standard. Further, although body mass index can serve as a surrogate measure of health risk, body mass index can both over- and underestimate adiposity and muscle content. Hence, recent obesity consensus guidelines recommend confirmation of excess adiposity through direct measurement of body fat, where available (5). Finally, sarcopenia co-occurring with obesity, identifiable via CT, warrants recognition as a distinct clinical entity due to the pathogenic interaction between the two and the increased risk of functional decline and metabolic disorders that surpasses the impact of either condition alone (6).

Using obesity as a comparator, the aim of this study was to investigate our hypothesis that there is a discrepancy with sarcopenia reporting, both with EHR diagnostic coding and with CT reports, versus the actual presence of reduced muscle mass on CT images.

Materials and Methods

Following institutional review board approval and in accordance with the Health Insurance Portability and Accountability Act, a retrospective review of the EHR was performed for 17 646 adult patients who consecutively underwent contrast-enhanced abdominopelvic CT between 2013 and 2018 at an academic institution. In the EHR, the frequency of sarcopenia and obesity was determined by using diagnostic codes (International Classification of Diseases, Ninth Revision [ICD-9] code 728.2/International Classification of Diseases, Tenth Revision [ICD-10] code M62.84, sarcopenia; ICD-9 code E66.01/ICD-10 code E66.09, obesity) and analysis of CT reports (search terms: “myopenia,” OR “myosteatosis,” OR “sarcopenia,” OR “obese,” OR “obesity”). Using axial CT images, the frequency of sarcopenia and obesity was determined with a previously validated open-source artificial intelligence (AI) tool that labels muscle and fat at the L3 level (Comp2Comp; https://github.com/StanfordMIMI/Comp2Comp), yielding a skeletal muscle index (which equals total muscle area divided by patient height squared, the latter of which was obtained from the medical record) and visceral adipose tissue area (Figure).

Example CT scans show skeletal muscle and visceral adipose tissue (VAT) area measurements in four patients with a similar body mass index (BMI), calculated as patient weight in kilograms divided by patient height in meters squared, at the L3 vertebral level, along with corresponding skeletal muscle index (SMI) and VAT T-scores.

Skeletal muscle index and visceral adipose tissue T-scores, a measure of the number of SDs from a healthy young individual of the same sex, were derived using published normative values established in healthy kidney donors. To provide benchmark measures of diagnosis rates, we compared the frequency of sarcopenia and obesity using the International Classification of Diseases (ICD) codes across 274 hospitals (Vizient; https://www.vizientinc.com).

Results

The average age of patients was 56.2 years ± 17.3 (SD). The Table shows the demographic characteristics, as well as the EHR frequency of sarcopenia and obesity at our institution and across 274 institutions, with comparison to the frequencies determined from our CT scans and previously reported in the general population.

Characteristics of the Study Population

In the EHR at our institution, sarcopenia was documented in 0.05% (nine of 17 646) of patients using ICD codes and mentioned for 0.005% (one of 17 646) CT reports, while obesity was diagnosed in 17.4% (3066 of 17 646) and 0.16% (28 of 17 646), respectively, of patients. On CT scans, using a threshold of T-score less than −2, sarcopenia was found in 28.5% (5030 of 17 646) of patients and obesity was found in 15.7% (2766 of 17 646). Coexisting sarcopenia and obesity was not documented (0 of 17 646) in the EHR, for both ICD codes and report mentions. On CT scans, sarcopenic obesity was identified in 5.7% (1008 of 17 646) of individuals.

Across 274 institutions, diagnostic codes for sarcopenia averaged 0.1% (range, 0.0%–0.1%) compared with 19% (range, 15%–24%) for obesity (7). The estimated community prevalence of sarcopenia is 10%–24%, and the prevalence of obesity in our county and in the United States ranges from 21% to 42% (12).

Discussion

Automated CT image analysis found that objective measurement of low muscle mass was three orders of magnitude higher than documented in the medical record with EHR coding or CT reports. The frequency of sarcopenia diagnosis in the EHR at our institution (0.05%) was comparable to that of discharge codes across 274 institutions (0.1%). In contrast, using the CT images, we found sarcopenia in 28.5% of patients, comparable to the prevalence estimates in the sarcopenia literature (6). Moreover, 5.7% of patients were identified as having sarcopenic obesity, a condition not otherwise documented in the EHR. These results underscore the opportunity to use automated AI tools for analysis of CT scans acquired during routine clinical care, without additional cost or radiation exposure to patients.

Our study had limitations. First, the CT metrics were derived from a single-site cohort, motivating further study in other care settings. Second, although we used visceral adipose tissue as an indicator of obesity due to its strong association with metabolic disease and mortality, incorporating other ectopic fat depots and total fat measurements could provide additional characterization of obesity. Third, manual verification of segmentation errors was not performed due to the study size. Although prior validation of automated measurements in this population has shown minimal errors, residual errors remain possible. Future studies may explore the specification of error thresholds given their implications for clinical decision-making.

In conclusion, this study provides compelling evidence that sarcopenia and sarcopenic obesity are underdiagnosed clinically in the EHR and in CT reports. Given that AI tools are now available to quantify body composition features such as low muscle mass and increased visceral adiposity, factors that stratify risk and may affect management for individual patients, there is a crucial need to explore real-time implementation of AI tools for body composition.