Attributable Perioperative Cost of Frailty after Major, Elective Noncardiac Surgery: A Population-based Cohort Study

This was a population-based cohort study conducted in Ontario, Canada (more than 14 million inhabitants), where universal physician and hospital health insurance (Ontario Health Insurance Plan) is provided to all residents. Healthcare data in Ontario are collected using standardized methods¹⁸  and stored at ICES, an independent research institute in Ontario, Canada. Datasets used for the study included: the Discharge Abstract Database (all hospitalizations); the National Ambulatory Care Reporting System (emergency and outpatient care); the Continuing Care Reporting System (details of long-term and respite care); the Ontario Drug Benefits Database (prescription drug claims for residents 65 yr or older); the Registered Persons Database (all death dates), Home Care Database (receipt of home-based health services) and National Rehabilitation System (receipt of inpatient rehabilitation). The dataset was assembled and analyzed by a trained data analyst (R.T.) after protocol registration (https://osf.io/qy9h8/). Because all data were routinely collected and de-identified, the need for ethical review was legally waived. The article is reported according to appropriate guidelines.^19,20 

We identified all Ontario residents aged 66 yr or older on the day of elective, major noncardiac surgery (an age range that reflects the older adult demographic and that allowed full capture of prescription medication data). Participants were identified using standardized methods to capture surgeries that occur during an elective hospital admission using established noncardiac surgical Canadian Classification of Interventions codes that represent surgeries of intermediate to high physiologic stress (see procedures in Supplemental Digital Content 1.0, https://links.lww.com/ALN/D152). These procedure codes have been used together extensively in previous studies.^21–25  To create a participant-level data set we identified the first eligible surgery for each participant from April 1, 2012, to March 31, 2018.

Frailty was identified using a multidimensional frailty index²⁶  validly ascertained from health administrative data using the recommended methods of Searle et al.²⁷  This approach ensures that the required 30 multidimensional deficits are available for all participants to allow calculation of a frailty index value between 0 (no deficits present) and 1 (deficits present in all domains) based on the accumulating deficits model of frailty (Supplemental Digital Content 2.0, https://links.lww.com/ALN/D152).²⁶  The index includes dimensions assessing medications, comorbidities (cardiovascular, pulmonary, endocrinologic, neurologic, rheumatologic, liver and renal comorbidities), cancer, dementia, socioeconomic status, need for supported living, substance abuse, dental, skin, hearing and vision issues, and receipt of supportive medical devices. For primary analyses, frailty was coded as present or absent based on a frailty index value of less than 0.21 versus 0.21 or greater, respectively.^26,28  The calculated frailty index value was also captured as a continuous variable for sensitivity analyses.

Our primary outcome was total health system costs in 365 days from surgery (starting on the day of surgery, expressed in 2018 Canadian dollars) paid by the Ministry of Health, the public insurer for all Ontario residents. Standard and validated case costing algorithms were used to capture all publicly funded direct and indirect costs (e.g., hospitalization, rehabilitation, nursing home care, medical diagnostics, pharmacy and provincially funded home-based health care),²⁹  accounting for each individual’s resource intensity weight. We also calculated the component costs: hospitalization (inpatient, outpatient, or ambulatory surgical costs), postacute care (rehabilitation, complex continuing care, long-term [i.e., skilled nursing facility] care, and home care services), physician billing (all physician service claims or capitated payments), and other (cancer care, dialysis, emergency department care, prescription medications). Out-of-pocket costs for patients (e.g., private physiotherapy or noninsured medical devices) were not captured. Secondary outcomes were costs accumulated at 30 and 90 days after surgery.

Variables that were postulated to be direct causal contributors to exposure, or present on the pathway between exposure and outcome, were not adjusted for in the primary analysis to avoid overadjustment bias.³⁰  Demographics (age, sex, neighborhood income quintile, rural vs. urban residence) were identified from the Discharge Abstract Database and from the Canadian Census. Standard methods were used to identify all Elixhauser comorbidities using International Classification of Diseases, Tenth Revision, codes from the Discharge Abstract Database in the 3 yr preceding surgery.³¹  Prescription medications were captured from the drug database. Acute care hospitalizations, emergency department visits, and total health system costs in the year before surgery were captured from hospital and emergency department records and using costing algorithms. Preoperative residence in a long-term care facility was identified from the Continuing Care Reporting System. Specific inclusion and parametrization of confounders in adjusted models are specified in the Analysis section.

This was a population-based study; we included all eligible members of the Ontario population. With approximately 20,000 eligible surgeries per year in Ontario, we estimated that mean 1-yr healthcare costs of one older person having surgery would approximate $15,000 Canadian dollar ± 5,200.³²  Because our primary objective required estimation of a stable multivariable linear regression model for log-transformed costs (to account for skew), the methods of Riley et al.³³  suggested a minimum sample size of 488.

Our approach to missing data was prespecified in our registered protocol. Because no exposure or outcome data were missing, their imputation was not required. Because neighborhood income quintile was missing for 0.4% of participants, we centrally imputed the most common value (the median [i.e., middle] income quintile).

Description of the study cohort was stratified by frailty status, with differences in characteristics quantified using absolute standardized differences.³⁴  All analyses were performed using SAS v9.4 for Windows (SAS Institute, USA), and 95% CIs were estimated to reflect uncertainty in applying our findings outside of our study population because our approach to inference was not based on a null hypothesis testing framework.

We estimated the unadjusted and adjusted association of frailty with total health system costs in the year after surgery using linear regression with log-transformation of the cost-dependent variable to account for the expected skew.³⁵  The exponentiated coefficient from this model represents a ratio of means, reflecting the relative difference in mean costs between those with and without frailty. In addition to a binary term for frailty, adjusted models included terms for sex (binary), age (restricted cubic spline with three knots), year of surgery (restricted cubic spline with three knots), and procedure (categorical for each procedure code). Generalized Estimating Equations accounted for clustering in hospitals using an exchangeable correlation matrix. The absolute attributable difference in mean cost was estimated using predicted cost outputs from our log-transformed linear regression model, with uncertainty estimated via a nonparametric bootstrapped standard error across 200 resamples with replacement (computational limitations with our population-based sample precluded use of greater replicates).³⁶  The association of frailty with costs at 30 and 90 days after surgery were estimated as described for 1-yr costs.

We conducted a series of prespecified sensitivity analyses. All sensitivity models included the same variables as the adjusted primary analysis, unless otherwise specified. First, to estimate the possible impact of comorbidity independent of frailty we additionally adjusted for each preexisting Elixhauser comorbidity as binary variables. Next, to explore the influence of each patient’s preoperative cost accumulation trajectory independent of frailty, we additionally adjusted for costs accrued in the year before the index surgery (parameterized as a restricted cubic spline with three knots). We also estimated the association of the frailty index with 1-yr costs as a continuous, nonlinear term using a fractional polynomial.³⁷  Because individuals with shorter survival may have less time to accumulate costs, we re-ran our primary analysis with the log of time alive as an offset term to allow calculation of the rate of cost accumulation adjusted for survival time.

We tested whether effect modification existed on the relative scale between frailty and prespecified potential effect modifiers. To do so, we included a multiplicative interaction term in the primary adjusted model between frailty and: (1) procedure category; (2) sex; and (3) age (accounting for potentially nonlinear effects using a three-knot restricted cubic spline). For categorical effect modifiers, we estimated the frailty–cost association at each level of the effect modifier variable. For continuous effect modifiers, we created a plot to visualize the estimated association across the range of recorded values for the effect modifier.

We identified 171,576 older patients having elective, major noncardiac surgery; 23,219 (13.5%) were identified with preoperative frailty. Patient characteristics by the presence of frailty are provided in table 1. Patients with frailty tended to have higher costs accumulated in the year preceding surgery and were more likely to be diagnosed with comorbidities. Patients without frailty presented more often for joint replacement surgery, whereas patients with frailty more often had major vascular or cancer-related surgeries. No required data were missing.

Total and component perioperative costs in the year starting on the day of surgery for patients with and without frailty are presented in table 2. The unadjusted ratio of means for costs at 1 yr in patients with frailty compared with those without frailty was 1.79 (95% CI, 1.76 to 1.83) and 1.53 (95% CI, 1.51 to 1.56) after adjustment for confounders. This was equivalent to an absolute mean-adjusted cost difference of $11,828 (95% CI, 11,805 to 11,850) attributable to the presence of preoperative frailty. Costs were higher for those with frailty within each component; postacute care costs had the largest relative increase (3.2-fold increase; table 2). Unadjusted and adjusted costs were also higher for those with frailty at 30 days (unadjusted ratio of means 1.42; 95% CI, 1.39 to 1.45; adjusted ratio of means 1.26; 95% CI, 1.25 to 1.27) and 90 days (unadjusted ratio of means 1.57; 95% CI, 1.54 to 1.61; adjusted ratio of means 1.37; 95% CI, 1.35 to 1.38) after surgery.

When preoperative frailty status was parameterized as a continuous variable, there continued to be a significant association with postoperative costs. As a continuous linear term, each 0.1 unit increase in the frailty index was associated with a relative increase in mean costs of 1.39 (95% CI, 1.37 to 1.41; fig. 1). As a nonlinear fractional polynomial, increasing frailty index score was associated with increasing costs across the recorded range of frailty index values, with an inflection point at a frailty score of 0.4 (fig. 1).

After additional adjustment for comorbidity, the adjusted frailty–cost association was attenuated (ratio of means 1.24; 95% CI, 1.22 to 1.26). A similarly attenuated association was found when we additionally adjusted for 1-yr preoperative costs (ratio of means 1.27; 95% CI, 1.26 to 1.29). Finally, 2,816 (12.1%) patients with frailty died within 365 days after surgery compared with 4,046 (2.7%) patients without frailty. Accounting for the rate of cost accumulation per time alive, the adjusted association between frailty and costs was larger than for the primary analysis (ratio of means 1.71; 95% CI, 1.68 to 1.73).

The association between frailty and perioperative costs was slightly greater for women with frailty (ratio of means 1.56; 95% CI, 1.54 to 1.59) compared with men with frailty (ratio of means 1.51; 95% CI, 1.49 to 1.53). The association between frailty and costs varied by surgical procedure and was most pronounced after total hip replacement (ratio of means 1.68; 95% CI, 1.64 to 1.72). The smallest association was estimated after pancreaticoduodenectomy (ratio of means 1.20; 95% CI, 1.13 to 1.29; fig. 2). Frailty was associated with the greatest relative increase in costs at younger ages. The association continuously decreased in a curvilinear fashion up to age 95 (fig. 3). For example, compared with those without frailty, patients aged 65 yr with frailty had a 1.69-fold (95% CI, 1.63 to 1.75) increase in costs, whereas patients aged 95 yr had a 1.41-fold (95% CI, 1.35 to 1.48) increase. To provide context for these differences based on postulated effect modifiers, the beta coefficient for each interaction term tested excluded the null value (0) from its 95% CI.

In this population-based cohort study of older patients undergoing elective, major noncardiac surgery, we estimate that a 1.5-fold increase in adjusted costs in the year after surgery is attributable to the presence of preoperative frailty. This association appears to be present across the first postoperative year with effect estimates being larger the farther from surgery costs were accrued. In sensitivity analyses adjusting for comorbidities or preoperative costs, the association of frailty with increased postoperative costs was maintained, although the effect size was attenuated to a 1.24-fold increase and a 1.27-fold increase, respectively. Further modifications of the frailty–cost association were also notable in younger patients and after certain procedures, especially total joint replacement. As the number of older people with frailty requiring surgery continues to increase, the additional costs required to care for such patients will need to be allocated in health system planning. Future research should aim to develop and evaluate approaches to help improve outcomes and reduce healthcare resource utilization for older adults with frailty.

Frailty is consistently associated with at least 2-fold adjusted increases in adverse clinical and patient-reported outcomes after surgery.^12,16,38,39  Therefore, it is not surprising that health system costs are also strongly associated with the presence of preoperative frailty. However, quantifying the relative and absolute increases in costs attributable to frailty using multidimensional and clinically relevant frailty exposure and robust health system costing methods is key to inform health systems as they increasingly face the rapid aging of our population and the growing number of older people with frailty presenting for surgery.^40,41  Based on our data, during 5 yr in a province with a population of 14 million, each of more than 23,000 elective surgeries for people with frailty cost more than $25,000 extra per case; this equates to an additional $600 million in health system spending. Although adjustment for confounders attenuated this cost difference, adjusted attributable costs still amounted to almost $275 million in additional health spending attributable to preoperative frailty. This means that, without adequate resourcing, surgical care for older people with frailty is likely to consume a growing portion of perioperative budgets over time.

In addition to adequate budgeting by payors and system administrators, ours and related data further help to inform considerations around the higher costs attributable to frailty. First, although limited by use of different frailty instruments and control for confounding, the relative increase estimated in our study is consistent with previous studies using the proprietary Adjusted Clinical Groups frailty defining diagnoses indicator, which have also demonstrated an approximately 1.5-fold increase.¹⁶  This suggests that the degree of relative cost increase attributable to frailty may be generalizable. Next, although we found increases in all aspects of postoperative health spending, postacute care costs were more than three times higher for people with frailty. Because postacute care appears to be a major driver of increased costs and nonhome discharge is identified by older surgical patients as a primary adverse outcome to avoid,^42,43  research and program development are urgently needed to help older people with frailty maintain independence after surgery and transition home effectively.⁴²  Although prehabilitation appears to be promising in this regard,⁴⁴  available high-quality trials specific to frailty fail to demonstrate efficacy without high levels of program adherence.^45,46 

Although we found significant associations between frailty and costs across all primary, secondary and sensitivity analyses, we also identified substantial modifiers of this association. Likely reflecting the increased demands for postacute care, the relative association of frailty and costs was greatest at a year after surgery and attenuated at postoperative days 30 and 90. Additionally, due to a higher mortality rate among people with frailty, our primary analysis may have underestimated the association of frailty with costs among those who survived the full postoperative year. When we accounted for time alive, the adjusted association of frailty with costs rose from a 1.5- to a 1.7-fold increase. This suggests that, if interventions are developed to improve clinical outcomes for surgical patients with frailty, they may not directly decrease costs. However, they could increase the value of perioperative care if added resources result in improved long-term patient outcomes.⁴⁷  Effect modification by age and surgical procedure largely reflected directional effects seen in the frailty–mortality association, with larger effect estimates identified at younger ages and after relatively physiologically less intense surgery like joint replacement.⁴⁸  Reasons for attenuated associations after more intensive surgeries like pancreaticoduodenectomy remain to be elucidated but could be related to factors present in patient selection that are not captured in routinely collected data, or the impact of frailty being diluted after procedures with higher baseline rates of morbidity. Ultimately, these data suggest that more granular approaches may be required to address health system costs across surgery types and the full perioperative journey.

Use of population-based data support generalizability to similar jurisdictions and health systems; however, more contemporary data would improve generalizability to current practice. Generalizability to substantively different systems also cannot be directly inferred. We used validated and reliable methods to identify our cohort, our exposure, and our outcome, and followed a pre-registered protocol. This should reduce misclassification and reporting bias. Although we were able to identify different component inputs to increased total costs, future research will be required to understand how adverse events and interventions may mediate cost accumulation. Future research will also be required to understand the out-of-pocket and informal caregiving costs incurred by patients and their families after surgery.

Frailty is associated with a significant increase in health system costs across the first postoperative year, resulting hundreds of millions in extra health system spending. However, the attributable costs vary based on patient and procedural characteristics. Future efforts to deliver perioperative care for the aging population will require health system planning, program development, and implementation strategies that consider the unique resource needs and trajectories of older patients with frailty.

This study was supported by ICES, which is funded by an annual grant from the Ontario Ministry of Health and the Ministry of Long-Term Care (Toronto, Canada). This document used data adapted from the Statistics Canada Postal Code^OM Conversion File, which is based on data licensed from Canada Post Corporation (Ottawa, Ontario, Canada), and/or data adapted from the Ontario Ministry of Health Postal Code Conversion File, which contains data copied under license from Canada Post Corporation and Statistics Canada (Ottawa, Ontario, Canada). Parts of this material are based on data and/or information compiled and provided by: Ministry of Health, Canadian Institute for Health Information (Ottawa, Ontario, Canada). The analyses, conclusions, opinions and statements expressed herein are solely those of the authors and do not reflect those of the funding or data sources; no endorsement is intended or should be inferred.

Dr. McIsaac receives salary support from The Ottawa Hospital Department of Anesthesiology Alternate Funds Association and a Research Chair from the University of Ottawa Faculty of Medicine (Ottawa, Ontario, Canada).

Dr. Hallet has received speaking honoraria from Ipsen (Mississauga, Ontario, Canada), Advanced Accelerator Applications (Millburn, New Jersey), Medtronic (Vancouver, British Columbia, Canada), and Bristol Myers Squibb (New York, New York). The other authors declare no competing interests.