Validation of a frailty index in patients undergoing curative surgery for urologic malignancy and comparison with other risk stratification tools

Abstract

Objective: To retrospectively validate and compare a modified frailty index predicting adverse outcomes and other risk stratification tools among patients undergoing urologic oncological surgeries.

Materials and methods: The American College of Surgeons National Surgical Quality Improvement Program was queried from 2005 to 2013 to identify patients undergoing cystectomy, prostatectomy, nephrectomy, and nephroureterectomy. Using the Canadian Study of Health and Aging Frailty Index, 11 variables were matched to the database; 4 were also added because of their relevance in oncology patients. The incidence of mortality, Clavien-Dindo IV complications, and adverse events were assessed with patients grouped according to their modified frailty index score.

Results: We identified 41,681 patients who were undergoing surgery for presumed urologic malignancy. Patients with a high frailty index score of >0.20 had a 3.70 odds of a Clavien-Dindo IV event (CI: 2.865-4.788, P<0.0005) and a 5.95 odds of 30-day mortality (CI: 3.72-9.51, P<0.0005) in comparison with nonfrail patients after adjusting for race, sex, age, smoking history, and procedure. Using C-statistics to compare the sensitivity and specificity of the predictive ability of different models per risk stratification tool and the Akaike information criteria to assess for the fit of the models with the data, the modified frailty index was comparable or superior to the Charlson comorbidity index but inferior to the American Society of Anesthesiologists Risk Class in predicting 30-day mortality or Clavien-Dindo IV events. When the modified frailty index was augmented with the American Society of Anesthesiologists Risk Class, the new index was superior in all aspects in comparison to other risk stratification tools.

Conclusion: Existing risk stratification tools may be improved by incorporating variables in our 15-point modified frailty index as well as other factors such as walking speed, exhaustion, and sarcopenia to fully assess frailty. This is relevant in diseases such as kidney and prostate cancer, where surveillance and other nonsurgical interventions exist as alternatives to a potentially complicated surgery. In these scenarios, our modified frailty index augmented by the American Society of Anesthesiologists Risk Class may help inform which patients have increased surgical complications that may outweigh the benefit of surgery although this index needs prospective validation.

Keywords: Frail elderly; Patient survival; Preoperative evaluation; Surgical outcomes; Urologic oncology.

Figure 1

Receiver operator characteristics (ROC) curve for mortality using our mFI in comparison to the existing parameters of predicting adverse outcomes. Our mFI had very poor sensitivity and specificity for predicting death in radical cystectomy (RC, Fig. 1.d C-statistic 0.574, p< 0.0005), fair sensitivity and specificity for predicting death in radical prostatectomy (RP, Fig. 1.a, C-statistic 0.760, p<0.0005), fair sensitivity and specificity for predicting death in nephroureterectomy (Neph-U, Fig. 1.c, C-statistic 0.753, p<0.0005), and poor sensitivity and specificity for predicting death in partial and radical nephrectomy (PN and RN, Fig. 1.b, C-statistic 0.698, p<0.0005). In all cases except RC, our 15-point mFI performed better than the ASA Risk Class Stratification System and the Charlson Comorbidity Index. For RC, the ASA Class outperformed the mFI with a C-statistic of 0.612 (p<0.0005) in comparison to the 15-point mFI that had a C-statistic of 0.574 (p<0.0005). Our 15-point mFI was superior to the 11-point CSHA-FI in all cases.

Figure 2

Receiver operator characteristics (ROC) curve for Clavien-Dindo IV outcomes using the mFI in comparison to the existing parameters of frailty. The mFI had poor sensitivity and specificity in radical prostatectomy (RP, Fig. 1.a, C-statistic 0.615, p<0.0005), very poor sensitivity and specificity in radical cystectomy (RC, Fig. 1.d, C-statistic 0.585, p< 0.0005), poor sensitivity and specificity in nephroureterectomy (Neph-U, Fig. 1.c, C-statistic 0.691, p<0.0005), and poor sensitivity and specificity in radical and partial nephrectomy (RN and PN, Fig. 1.b, C-statistic 0.646, p<0.0005). However, the mFI equaled or surpassed the Charlson Comorbidity Index or ASA Class Risk stratification in RN, PN and Neph-U. In RP, the ASA Class outcompeted the mFI with a higher C-statistic of 0.623 in comparison to 0.615. In RC, the ASA Class also outcompeted the mFI with a higher C-statistic of 0.612 in comparison to 0.585. The 15-point mFI was superior to the 11-point CSHA-FI in all the comparisons.

Figure 3

A comparison of different risk stratification tools with the modified frailty index in our multivariate model. The parameters measured to assess the different models were Akaiki information criteria (AIC) and the C- Statistic. A low AIC indicates better goodness of fit while a higher C-statistic value indicates an optimized model with both good sensitivity and specificity for a given outcome. The outcomes assessed were mortality (Figure 3.a) and Clavien-Dindo IV complications (Figure 3.b). The modified frailty index had fair sensitivity and specificity (C-statistic for mortality 0.66, for Clavien Dindo IV Complications 0.72) while maintaining a low AIC (AIC for mortality= 2400.6, AIC for Clavien Dindo IV Complications=8371.6) although the ASA Class Risk groupings outperformed it in both outcomes with an equal or higher sensitivity and specificity (C-statistic for mortality 0.67, for Clavien Dindo IV Complications 0.72) and lower AIC (AIC mortality=2406.1, AIC Clavien Dindo IV Complications =8345.1). However, when the ASA Class Risk Group and the mFI were combined, it was superior in all regards with lowest AIC (Mortality=2372.7, Clavien Dindo IV Complications =8321.4) and the highest C-statistic (Mortality 0.71, Clavien Dindo IV Complications = 0.77).