Preoperative Frailty Assessment: Reply

We compared the Modified Frailty Index and Hopkins Frailty Score for predicting duration of hospitalization.¹  Johnson et al.²  and Darvall et al.³  note that the scores we used differ from those recommended by the 2012 frailty operative definition consensus conference.⁴  The consensus conference used a modified Delphi process to consolidate expert opinion and generate a consensus definition for frailty. Although participating experts agreed that frailty is multidimensional, they could not agree on an operational definition, possibly because supportive data for individual frailty components are lacking. The assessment tools we used assess single—but different—frailty dimensions. In this respect, they were similar to most perioperative frailty assessment tools that are also unidimensional. For example, a recent systematic review of reported frailty instruments describes 51 tools, of which only two are multidimensional.⁵ 

Johnson et al.²  comment that the Hopkins Frailty Score only measures the physical aspects of frailty. However, our study was not to develop or test a specific multidimensional frailty score; instead it was designed to compare the clinical performance of the two most commonly used measures. Hopkins Frailty Score closely mimics the commonly used Fried Index and is validated for noncardiac surgical patients. A recent systematic review and meta-analysis identified the Fried Index and related modifications to be the most prevalent preoperative frailty measure among the 35 tools they evaluated.⁶  Despite its lack of multidimensionality, the Fried Index is recommended by the American College of Surgeons as a tool for assessment of frailty among geriatric surgical patients.⁷ 

The other frailty tool we used, the Modified Frailty Index, was developed by selecting 11 of the variables collected by the National Surgical Quality Improvement Program. The selected variables closely match those in the Canadian Study of Health and the Aging-Frailty Index, a well regarded deficit accumulation model of frailty.⁸  Among the 11 variables constituting the Modified Frailty Index, nine are comorbidities. The other two are impaired sensorium and functional status, which measures cognitive domain and physical performance. We do acknowledge the observations of Johnson et al.²  and Darvall et al.³  that the Modified Frailty Index is largely a measure of comorbidity rather than a multidimensional measure of frailty. We nonetheless considered the index because it is among the most commonly reported frailty scores within the deficit accumulation paradigm. In fact, a recent meta-analysis assessing frailty in noncardiac surgical population evaluated the Modified Frailty Index because it was the most commonly used perioperative measure.⁹  Specifically, the authors identified 32 studies that used the Modified Frailty Index, most published after 2015.

Referring to the work of Searle et al.,¹⁰  Darvall et al.³  state that at least 30 deficits must be included to construct an accurate deficit accumulation frailty index. We agree that more information increases precision of frailty measures, but measures that assess 30 or more deficits are relatively impractical and seem unlikely to be broadly adopted. Furthermore, such elaborate measures may not be necessary. Although Searle et al.¹⁰  stated that 30 to 40 deficits is optimal, they also conclude that models with at least 10 deficits are stable.

We agree with Johnson et al.²  that duration of hospitalization is influenced by various variables including social factors. Nevertheless, it is primarily driven by various disease-related factors and to some extent by hospital-related factors. Hospital duration is therefore a widely reported outcome in frailty studies, including ones that used the Modified Frailty Index and the Fried Index. Johnson et al.²  comment that using historic trends to calculate length of stay downwardly biases accuracy. However, our length-of-stay data were obtained from 2010 to 2015, which is well within the accepted time frame of temporal transposability. Johnson et al.²  suggested that an alternative study design adjusting for procedure as a covariate in a contemporaneous cohort might have better performance. However, choosing a contemporaneous cohort for calculating expected length-of-stay from a different hospital setting would not be preferable because hospital-related factors play an important role in determining length of stay. Even if our estimates of expected length-of-stay were consistently high, it would not influence the relative ability of each measure to predict hospital duration. In any case, we compared the two frailty measures within each patient, thereby avoiding bias.

We agree with Johnson et al.²  that dichotomizing continuous variables diminishes statistical power—and therefore did not. As specified in the text, “Patients with a Hopkins Frailty Score of 3 or more are classified as frail. Among various cutoffs used to designate frailty based on the Modified Frailty Index, the score of 4 or higher out of 11 is most commonly reported. However, both scores were considered as continuous variables for our analyses.” As specified, dichotomized cohorts were therefore presented in the tables only to facilitate interpretation.

In summary, our results support the 2012 frailty consensus statement by showing that the Hopkins Frailty Score and the Modified Frailty Index, both of which lack multidimensionality, poorly predict the duration of hospitalization and complications after noncardiac surgery. It will be interesting to see whether multidimensional measures predict better.

The authors declare no competing interests.