This study shows that EFS-determined vulnerability is a predictor of postoperative in-hospital 4AT scores ≥4 in older non-ICU patients undergoing elective surgery. Among the EFS domains, the strongest associations with in-hospital 4AT scores ≥4 were requirements for assistance with activities of daily living, presence of incontinence, difficulty with timed get up and go, and forgetting to take medications. EFS criteria for vulnerability are predictors of POD in older surgical patients undergoing lower-risk procedures.
In general, most frailty instruments demonstrate associations between frailty and poorer outcomes [12]. For instance, frailty instruments add predictive value for death, new disability, and LOS after major elective surgery [13]. However, surgical outcome studies vary considerably in both type of frailty instrument used and the frailty incidence detected. Unfortunately, there are few head-to-head comparisons of frailty instruments in terms of their ability to predict POD. Data from a meta-analysis that compared EFS and Fried criteria suggested stronger associations for POD with the Fried criteria [4]. In studies providing area under the curve (AUC) data, models report an AUC of 0.695 [14] using the modified Fried criteria, whereas a study using the Groninger criteria reported an AUC of 0.89 [15]. Our bivariate analysis adjusted for Elixhauser comorbidity score and EFS had an AUC of 0.796. Our multivariate analysis adjusted for age, ASA category, Elixhauser comorbidity score, and EFS had an AUC of 0.833. Both of our models that incorporated EFS demonstrated excellent predictive capability for delirium using the EFS ≥ 6 cutoff.
Our study population varied considerably from those in earlier reports. Patients requiring ICU admission or urgent/emergent surgery were excluded. In addition, we included a broad range of surgical specialties. Emergency/urgent surgery, ICU admission, and procedures with high cardiac risk are all strong risk factors for delirium [16]. Their elimination from the study population accounts for the lower delirium case index. Our power calculations assumed a higher incidence rate than we observed. However, the observed odds ratio was higher, which is consistent with our findings that EFS-determined vulnerability is associated with 4AT ≥4. The lower case index and its associated issue of power is likely important. However, the direction of the frailty effect was as expected. Longer length of hospital stay was associated with 4AT ≥4 and higher Elixhauser comorbidity score, but it did not affect the association between POD and frailty. In any case, the strong association that we found emphasizes that use of EFS criteria still gives strong predictive value for POD in older surgical patients undergoing lower-risk procedures.
We wish to drill down on the frailty-delirium relationship by dissecting which EFS items are most closely associated with delirium. On the one hand, the questions on the EFS are not granular. But overall, most EFS domains showed a trend toward predicting delirium. This trend probably accounts for the strong relationship between EFS-detected vulnerability and delirium. The EFS domains that focused on function and mobility had strong associations with delirium and are consistent with those in the literature [16]. Urinary incontinence and delirium might be linked via need for anticholinergic administration or underlying cognitive dysfunction. However, chi-square analysis of anticholinergics with and without incontinence did not show statistical significance. Incontinence is also associated with presence of underlying neurologic disease. However, chi-square analysis of neurologic comorbidities from the Elixhauser score (paralysis, dementia, psychosis) with and without incontinence showed no significance. Thus, the association of the incontinence EFS domain with delirium occurs via some other mechanism.
Additional clinical importance of our study comes from our 30-day mortality predictor analysis (Supplemental table). In multivariate analysis, only postoperative 4AT scores ≥4 were predictive of 30-day mortality. The fact that only four deaths occurred within 30 days of surgery limits the generalizability of this result and accounts for the wide confidence intervals. Nonetheless, future studies on preventing delirium with interventions focused on delirium risk factors, such as frailty, may be important for decreasing 30-day mortality even in older populations undergoing low-risk surgery.
Strengths and limitations
Here we studied pragmatic frailty and delirium assessments. Both the EFS and 4AT are easily administered and feasible in preoperative surgical clinics and on surgical wards. Full EMR datasets were analyzed, providing opportunities for implementation of EMR-driven quality improvement dashboards. All types of elective non-ICU–requiring surgeries were included, giving a broader base to our understanding of frailty–delirium relationships. Nevertheless, some limitations must be considered. The study was retrospective in an in-hospital setting. Total anesthesia time is an imperfect indicator of surgical stress as some surgeries may take disproportionately longer, but are not necessarily more invasive. POD was assessed in a clinical setting, not up to the gold standard used in a research setting, with an implicit assumption of patients not having surgery-related POD after hospital discharge. Mild cognitive dysfunction may have been underrecognized as the clock draw is limited in its recognition of very mild dementia [17].
Not all wards accepting postoperative patients performed the 4AT and not all surgical clinics performed the EFS assessment preoperatively which limited our sample size and generalizability. The targeted sample size in power evaluation was based on feasibility determined by the past patient volume, but POD event rate in the evaluation was assumed according to general estimates and was too high in light of the lower surgical risk in the population being studied. The study had limited sample size resulting in a small number of cases (number of patients with 4AT score ≥ 4). We were careful in our analyses not to overfit the regression model for a small sized sample, for example, by conducting the adjusted analysis in a sequential way where we first adjusted for one relevant covariate at a time only (model 1 to 3 in Table 2). To address small sample bias in logistic regression analysis, we did use penalized likelihood approach [10] for our analyses. The limited sample size and lower number of events did not allow for extensive multivariable modeling; therefore, residual confounding cannot be ruled out. The small sample size also resulted in less precise association estimate, as apparent in the wide 95% confidence intervals for the odds ratio estimate. However, the results are in keeping with previous meta-analysis results. It is also worth noting that the OR estimate of 3.5 associated with EFS ≥ 6 from the multivariable model reported in Table 2 may represent a weighted average of association levels in the range of higher EFS scores observed in our study. In an exploratory analysis using a two-linear-segment logistic regression model, the OR estimate was 1.0 (0.6 to 1.5) per 1 point increase in EFS score between score of 0 and 5, and 1.8 (1.2 to 2.9) per 1 point increase in EFS score ranged from 6 to 9, the upper limit of EFS score observed in our sample (data not shown). Larger datasets from additional research will be needed to analyze the association using EFS as an ordinal variable that it is, and further examine the association with POD beyond the EFS range observed in our data.
