Abstract
BACKGROUND: The appropriate timing of hip fracture surgery remains a matter of debate. We sought to estimate the effect of changes in timing policy and the proportion of deaths attributable to surgical delay.
METHODS: We obtained discharge abstracts from the Canadian Institute for Health Information for hip fracture surgery in Canada (excluding Quebec) between 2004 and 2012. We estimated the expected population-average risks of inpatient death within 30 days if patients were surgically treated on day of admission, inpatient day 2, day 3 or after day 3. We weighted observations with the inverse propensity score of surgical timing according to confounders selected from a causal diagram.
RESULTS: Of 139 119 medically stable patients with hip fracture who were aged 65 years or older, 32 120 (23.1%) underwent surgery on admission day, 60 505 (43.5%) on inpatient day 2, 29 236 (21.0%) on day 3 and 17 258 (12.4%) after day 3. Cumulative 30-day in-hospital mortality was 4.9% among patients who were surgically treated on admission day, increasing to 6.9% for surgery done after day 3. We projected an additional 10.9 (95% confidence interval [CI] 6.8 to 15.1) deaths per 1000 surgeries if all surgeries were done after inpatient day 3 instead of admission day. The attributable proportion of deaths for delays beyond inpatient day 2 was 16.5% (95% CI 12.0% to 21.0%).
INTERPRETATION: Surgery on admission day or the following day was estimated to reduce postoperative mortality among medically stable patients with hip fracture. Hospitals should expedite operating room access for patients whose surgery has already been delayed for nonmedical reasons.
In Canada, hospitals admit 30 000 older adults with hip fracture annually.1 These patients face an increased risk of death, with up to 5% of women and 10% of men dying within 30 days.2,3 It is generally accepted that early operative intervention improves survival by reducing patients’ exposure to immobilization and inflammation.4 In 2005, the federal, provincial and territorial governments established a benchmark of 48 hours from admission for 90% of hip fracture surgeries to prevent potentially harmful delays.5 However, delays to hip fracture surgery remain common.6 Patients who are medically stable at presentation may have to wait until a surgeon or an operating room becomes available.7,8
There has been considerable debate about the point at which delaying hip fracture surgery for nonmedical reasons worsens mortality.9–25 This uncertainty leads to prioritization without benefit to the patient or underuse of expeditious surgery that could prevent deaths. Some have argued that understanding the effects of policy change should guide reorganization of operating room resources26 and prioritization in the presence of competing demand.7,27–29 In this paper, we offer 2 new estimates: the effect of possible changes in surgical timing policy in the same population of patients, and the proportion of in-hospital deaths attributable to surgical delays.
Methods
Study approach
Using population-based data, we contrasted the risk of in-hospital death that would be expected (i.e., marginal risk30) if all patients who are medically stable at presentation were to undergo surgery on the day of admission, on inpatient day 2, on inpatient day 3, or after inpatient day 3. We obtained these risks by stratifying observations on confounders identified from an evidence-informed causal diagram,31,32 and weighting observations with the inverse propensity score of surgical timing for their respective strata.33 We then combined the weighted observations across strata to construct equal-sized samples, each representing the same patient population treated on a certain day (Appendix 1, available at www.cmaj.ca/lookup/suppl/doi:10.1503/cmaj.171512/-/DC1).
Study population
The Canadian Institute for Health Information provided discharge abstracts for all patients who underwent hip fracture surgery between Jan. 1, 2004, and Dec. 31, 2012, except for those in Quebec.22 We combined multiple abstracts related to transfers to account for time spent at nonsurgical sites.34 We excluded patients with conditions that could delay hip fracture surgery;35 preoperative intensive care unit admission; more than 9 preoperative days (inconsistent with urgent nature of the procedure); surgery in a hospital with an annual volume of fewer than 24 hip fracture surgeries; or invalid discharge date. We identified abstracts with medical reasons for delay using diagnosis and procedure codes for anemia, anticoagulation, volume depletion, electrolyte imbalance, uncontrolled diabetes, uncontrolled heart failure, acute cardiac arrhythmia or ischemia, acute chest infection or exacerbation of a chronic chest condition.36
Outcome
The outcome was any death within 30 inpatient days after surgery, reported per 1000 surgeries.37 The accuracy of in-hospital death data has been validated previously.38 We focused on in-hospital deaths to isolate the acute phase of hip fracture care, because variation in quality and continuity of care after discharge from hospital may lead to exposure to risk factors that are not related to the timing of surgery. Hospital stays after 30 days were deemed nonacute.39
Exposure
The exposure was the timing of surgery, grouped as the day of admission (reference), on inpatient day 2, on inpatient day 3, and after inpatient day 3.40 These groups represent natural timing alternatives, because group membership was governed by the process of booking time in the operating room. After making the decision to operate, surgeons add patients’ names to the list of procedures to be carried out within 48 hours, or earlier if the patient has already been delayed by admission late in the day or transfer.7 Hospital managers book patients in the order of their addition to the list within the requested urgency category. Most hospitals do not offer evening surgery, and booking may not take place until the next morning if patients arrived after hours). Therefore, booking practices suggest access to surgery is an event occurring on a daily scale, rather than an hourly scale.
Selection of confounders
We used a causal diagram to select confounders, conditioning on which would be sufficient to render timing and mortality independent.32 Figure 1 shows known dependencies among factors that influence the timing of surgery and the occurrence of death, either directly or through a chain of dependencies.48 Using Figure 1, we identified the following factors: treatment era, hospital type, procedure type, age at admission, prefracture health status and surgical readiness (Appendix 1).41,49
Stratification
We constructed 64 (= 2 × 2 × 2 × 2 × 4 × 1) multifactor strata on the basis of treatment era (2004–2007, 2008–2012), hospital type (teaching, community), procedure type (fixation, arthroplasty), age at admission (65–84 yr, ≥ 85 yr) and prefracture health status (admitted from home without comorbidity, admitted from home with comorbidity or with home care services, admitted from a long-term care facility, or admitted from elsewhere) (Supplemental Table 1, Appendix 2, available at www.cmaj.ca/lookup/suppl/doi:10.1503/cmaj.171512/-/DC1). We considered the following comorbidities: cardiac dysrhythmias, chronic obstructive pulmonary disease, diabetes mellitus, heart failure, hypertension and ischemic heart disease (acute and chronic), identified by diagnostic codes from all hospital discharge abstracts in the year before the index admission. Needham and colleagues50 reported substantial agreement between comorbidities in Canadian Institute for Health Information discharge abstracts and hospital charts. Surgical readiness contributed 1 category, because we excluded patients whose discharge abstracts showed medically appropriate reasons for delay.
Statistical analysis
We estimated cumulative incidence of mortality, accounting for the rate of live discharge.51 Postoperative stays were treated as right-censored observations if they exceeded 30 days or if they ended with transfer to another acute care facility or with live discharge within 1 day after the surgery.52 We estimated the marginal risk of death as a population average of observations weighted by inverse propensity score of surgical timing, calculated as the proportion of patients with a certain timing of surgery within their respective strata (Appendix 1). We estimated risk differences and odds ratios relative to surgery on the day of admission using the respective marginal risks.53 Using the risk estimates for surgery done within 2 inpatient days and for surgery done later, we calculated the proportion of deaths that could be attributed to delaying surgery until after inpatient day 2, assuming all other contributing factors were distributed as in the study population.49 No outcome, exposure or confounder data were missing.
In sensitivity analyses, we compared stratification-based and model-based estimates.54 The stratification used fewer categories than available in our data for age, treatment year, comorbidity and hospital type to ensure a sufficient number of events per stratum. The variables for each model were identical to those in the stratification, but the number of categories for each variable corresponded to the observed data. For example, we entered a separate indicator variable for each comorbid condition into regression models. We performed all model-based analyses using the Stata “teffects” package, allowing for intra-hospital correlation among observations.55 We used the VanderWeele–Arah method to assess whether an unmeasured confounder could explain the risk difference between timing alternatives.56
Ethics approval
The University of British Columbia Behavioural Research Ethics Board approved this study.
Results
Characteristics of the study population
A total of 195 253 discharge abstracts were available for 154 389 patients treated at 188 hospitals between 2004 and 2012. After exclusions, the study population consisted of 139 119 patients aged 65 years or older who underwent surgery for nonpathologic first-time hip fracture at 38 teaching hospitals and 106 community hospitals (Figure 2). Most of these patients were women (103 405 [74.3%]), and almost half were 85 years or older (63 786 [45.8%]). Just more than half the surgeries (72 285 [52.0%]) were done because of a transcervical fracture, and 83 643 (60.1%) involved fixation (Table 1).
Surgical timing was distributed unevenly: 32 120 (23.1%) patients underwent surgery on the day of admission, 60 505 (43.5%) on inpatient day 2, 29 236 (21.0%) on day 3 and 17 258 (12.4%) after day 3, and surgical timing varied significantly within the strata (p < 0.001; Supplemental Figure 1, Appendix 2). Patients who underwent surgery soon after admission were less likely to have been admitted from home with comorbidities, were less likely to have been transferred, and were more likely to undergo fixation, with the percentage of patients who were transferred increasing and the percentage of patients with fixation declining as the time to surgery increased (Table 1).
By day 30 after surgery, 6371 (4.6%) of the patients had died, and 89 782 (64.5%) had been discharged alive. For 17 336 (12.5%) of the patients, the postoperative stay was longer than 30 days, and for 25 630 (18.4%), there was another censoring event. The mortality varied across the strata from 11.2 to 138.7 deaths per 1000 surgeries.
In-hospital death by timing of surgery
There were 3903 deaths after 92 625 surgeries done on the day of admission or inpatient day 2 (42.1 deaths per 1000 surgeries, 95% confidence interval [CI] 40.8 to 43.4), and 2468 deaths after 46 494 surgeries done on a later day (53.1 deaths per 1000 surgeries, 95% CI 51.0 to 55.1). The cumulative 30-day mortality was 48.9 deaths per 1000 surgeries done on the day of admission and 48.0 deaths per 1000 surgeries done on inpatient day 2 (Figure 3). For surgery carried out later, the mortality was significantly higher: 57.0 deaths per 1000 surgeries done on inpatient day 3 and 69.1 deaths per 1000 surgeries done after inpatient day 3.
Risk difference for timing alternatives
Weighting by the inverse propensity score resulted in 4 hypothetical samples of equal size, with identical distribution of the stratification factors but distinct timing of surgery (Figure 4). Table 2 shows the risks of in-hospital death that would be expected if all patients in the study were to undergo surgery on a certain day: 43.3 (95% CI 40.9 to 45.6) deaths per 1000 surgeries if all were done on the day of admission, 42.6 (95% CI 41.0 to 44.3) deaths per 1000 surgeries if all were done on inpatient day 2, 49.0 (95% CI 46.5 to 51.6) deaths per 1000 surgeries if all were done on inpatient day 3, and 54.2 (95% CI 50.8 to 57.7) deaths per 1000 surgeries if all were done after inpatient day 3.
Undergoing the procedure on inpatient day 2 rather than the day of admission did not change the risk of death for patients: risk difference −0.6 (95% CI −3.5 to 2.2) deaths per 1000 surgeries. However, if all surgeries were done on inpatient day 3 rather than the day of admission, there would be an additional 5.8 (95% CI 2.3 to 9.2) deaths for every 1000 surgeries, and the number of deaths would increase further, to 10.9 (95% CI 6.8 to 15.1) deaths for every 1000 surgeries, if all surgeries were done after inpatient day 3 (Table 2).
Relative to surgery on the day of admission, the marginal odds ratios were 0.98 (95% CI 0.92 to 1.05) for surgery done on inpatient day 2, 1.14 (95% CI 1.05 to 1.23) for surgery on inpatient day 3, and 1.27 (95% CI 1.16 to 1.38) for surgery after inpatient day 3 (Table 2). The estimates were similar for different specifications of the model-based analysis (Supplemental Table 3, Appendix 2). Among the 7183 patients who would be expected to die if all surgeries were delayed beyond inpatient day 2, 16.5% (95% CI 12.0% to 21.0%), or 1221 deaths, could be attributed to not undergoing surgery earlier.
Effect of unmeasured confounding
We also considered the effects of an unmeasured confounder. For example, evening surgery might be 20% more prevalent among surgeries done on the day of admission than among those after inpatient day 3, and might increase mortality by, say, 54.2 deaths per 1000 surgeries. The latter figure is artificially high because it equals to the mortality for surgery after inpatient day 3. This increase implies that mortality almost doubles for evening surgery, if mortality for patients with daytime surgery equals that for operations done on the day of admission; that is, 42 deaths per 1000 surgeries. We calculate that such unmeasured confounding would introduce a bias of 10.8 (= 54.2 × 0.2) deaths per 1000 surgeries, and therefore our estimate for the risk difference should be reduced to 0 (95% CI −4.1 to 4.1) deaths per 1000 surgeries.56 Alternatively, if evening surgery were to increase the risk of death by 20.9 deaths per 1000 surgeries, then a bias of 4.2 (= 20.9 × 0.2) would reduce the estimate to 6.7 (95% CI 2.3 to 10.8) deaths per 1000 surgeries. We therefore conclude that a single unmeasured confounder could produce the observed mortality differences only if it increased the risk of death by a factor of 2. It seems unlikely that a single unknown factor could have an effect sufficiently large to account, on its own, for the observed difference in mortality between the study groups.
Interpretation
We estimated the extent to which in-hospital mortality might change if the timing of hip fracture surgery had been different for a given patient population. We projected an additional 11 deaths for every 1000 surgeries if all patients considered in this study had undergone the operation after waiting 3 days or more, relative to surgery on the day of admission. The proportion of in-hospital deaths attributable to surgical delays beyond inpatient day 2 was estimated at 16.5%.
Lewis and Waddell12 concluded that considerable variation in practice and inconsistent evidence leave uncertainty about the optimal timing of hip fracture surgery. Lizaur-Utrilla and colleagues13 argued that there is no single timing of hip fracture repair that can be considered optimal for all, because of heterogeneity among patients, their injuries and their care needs. In the current study, we were concerned with the effect of changes in the timing policy rather than with the etiological question of whether delays worsen mortality. We compared expected mortality for timing alternatives if they had been implemented for the same patient population. Our projections refer to the total effect of the timing alternatives; whether postoperative complications might explain the differences in mortality across these timing alternatives requires further investigation.9
Several authors have acknowledged their failure to address imbalance between timing groups in terms of baseline variables that might influence outcomes.31,59–61 Therefore, inconsistent findings may result from differences between various surgical timing groups.62 We used an evidence-informed causal diagram to justify the selection of variables that would be sufficient to control for a spurious association between timing of surgery and mortality. However, there is still potential for unmeasured confounding, because the causal diagram includes only known factors and dependencies. Although our sensitivity analysis suggested that a single confounding factor could conceivably account for the between-group difference in mortality, such a factor would need to double the risk of postoperative death.
Limitations
We used administrative data, which might have led to misclassification of medical delays36 and underreporting of comorbidity.22 In particular, the available data omit renal disease, cerebrovascular accident and dementia, which may influence in-hospital mortality through increased risk of complications. Therefore, some observations in the group “admitted from home without comorbidity” might have been misclassified. However, the percentage of observations in this group was similar across the timing groups, and the percentage of “admitted from home with comorbidity” increased with time to surgery. The study population included only patients who underwent surgery; therefore, our analysis does not account for deaths that occurred before the surgery could be done.63
We studied delays that occurred after admission to hospital; data on time between injury and arrival at the hospital and in the emergency department were not available.64 It is possible the overall time from injury to surgery, and therefore exposure to immobilization and inflammation, was similar for inpatient day 1 and 2. We were unable to differentiate between surgeries done during and after working hours, because the data on timing of surgery on an hourly basis were not available. Booking surgery to occur after hours would reduce time to surgery, but it might also worsen mortality, because of reduced staffing and surgical team fatigue. Given strategies to address surgical delays that include increasing after-hours surgery, it will be important to study whether this approach produces better outcomes than waiting until the next day.
We did not differentiate between teaching hospitals of various sizes or rural and urban community hospitals; therefore, unobserved variation in care delivery across hospital types might have influenced both the timing of surgery and mortality. Finally, prefracture health status was characterized by a combination of comorbidity and preadmission residence.44 Although preadmission residence reflects health care needs, local supply may also influence admission to a long-term care facility.65
Conclusion
Our findings allow for the inference of a critical point for the timing of hip fracture repair. We suggest that all medically stable older adults with hip fracture undergo surgery on the day of their admission to hospital or the following day. This approach places the emphasis of managerial efforts on expediting operating room access for patients whose surgery might be delayed for nonmedical reasons.
Acknowledgements
The authors are grateful to Peggy Robinson, ELS, for presubmission manuscript editing, and to Duncan Hunter, Anna Chudyk and Christopher Filliter for useful feedback.
Footnotes
* The Canadian Collaborative Study of Hip Fractures: Lauren Beaupre, Eric Bohm, Michael Dunbar, Donald Griesdale, Pierre Guy, Edward Harvey, Erik Hellsten, Susan Jaglal, Hans Kreder, Lisa Kuramoto, Adrian Levy, Suzanne N. Morin, Katie J. Sheehan, Boris Sobolev, Jason M. Sutherland and James Waddell.
Visual abstract available at www.cmaj.ca/lookup/suppl/doi:10.1503/cmaj.171512/-/DC2
Competing interests: Boris Sobolev, Pierre Guy and the Collaborative report receiving grants from the Canadian Institutes of Health Research related to this work. Pierre Guy reports receiving grants from the Natural Sciences and Engineering Research Council of Canada, the Canadian Foundation for Innovation and the British Columbia Specialists Services Committee for work regarding hip fracture care, outside the submitted work. He has also received fees from the BC Specialists Services Committee (for a provincial quality improvement project on redesign of hip fracture care) and from Stryker Orthopaedics (as a product development consultant), outside the submitted work. He is a board member and shareholder in Traumis Surgical Systems Inc. and a board member for the Canadian Orthopaedic Foundation. He also serves on the speakers’ bureaus of AOTrauma North America and Stryker Canada. Suzanne Morin reports research grants from Amgen Canada, outside the submitted work. Jason Kim and Lisa Kuramoto report receiving grants from the Canadian Institutes of Health Research, during the conduct of the study. No other competing interests were declared.
This article has been peer reviewed.
Contributors: All authors contributed to the conception and design of the study. In addition, Boris Sobolev, Pierre Guy, Katie Jane Sheehan, Lisa Kuramoto and Jason Kim contributed to the acquisition and analysis of data. All authors contributed to the interpretation of the data. Boris Sobolev and Lisa Kuramoto drafted the manuscript. All authors revised it critically for important intellectual content, gave final approval of the version to be published and agreed to be accountable for all aspects of the work.
Funding: This research was funded by the Canadian Institutes of Health Research (MOP 133629). This funder has no role in the design of this study, execution, analyses, data interpretation or decision to submit results for publication.
Data sharing: We studied patient records that were anonymized and de-identified by a third party, the Canadian Institute for Health Information, an organization that provides researchers access to data on Canadian residents. Data are available from the Canadian Institute for Health Information for researchers who meet the criteria for access to confidential data.
- Accepted June 18, 2018.