Effect of early surgery after hip fracture on mortality and complications: systematic review and meta-analysis

Eligibility criteria

Studies fulfilling the following criteria were eligible for inclusion: target population consisting of patients 60 years of age or older who underwent surgery for a low-energy hip fracture, evaluation of preoperative surgical delay, consideration of all-cause mortality as an outcome and prospective design. We imposed no language restrictions.

Identification of studies

We used multiple strategies to identify potentially eligible studies. With the help of a professional librarian, we searched the electronic databases MEDLINE and EMBASE for relevant articles in any language that were published up to and including Feb. 8, 2008. The complete search strategies are shown in Appendix 1 (available at www.cmaj.ca/cgi/content/full/cmaj.092220/DC1). One reviewer (N.S.) also hand-searched the archives of annual meetings of the Orthopaedic Trauma Association (1996–2007), the International Society of Orthopaedic Surgery and Traumatology (2003–2007), the Canadian Orthopaedic Association (2006–2007), the European Federation of National Associations of Orthopaedics and Traumatology (2005–2007), the Mid-America Orthopaedic Association (2005–2007), the Piedmont Orthopedic Society (2005–2007), the Association of Bone and Joint Surgeons (2005–2007) and the American Academy of Orthopaedic Surgeons (2006–2007) for any relevant unpublished literature. Additional strategies to identify studies included consultation with experts, a manual review of the reference lists of articles that fulfilled our eligibility criteria and use of the “related articles” feature in PubMed for any study that met our criteria.

Screening and assessment of eligibility

One reviewer (N.S.) screened the titles and abstracts of the studies from the electronic search to identify all citations that might contain the comparison of interest. Two reviewers (N.S., S.S.) independently evaluated these studies, as well as studies identified by hand-searching of reference lists and abstracts from meetings, to determine final inclusion (Figure 1).

• Download figure
• Open in new tab
• Download powerpoint

Figure 1: Flow of studies through the systematic review.

Disagreements were resolved through a consensus process that required the reviewers to discuss the rationale for their decisions and come to an agreement. We recruited additional reviewers with an epidemiologic background and competence in various languages to apply the eligibility criteria to all non-English papers (one article in French, eight in German, three in Dutch and two in Hebrew). Study selection was not blinded, as blinding has been shown to have no significant statistical or clinical effect on the final results of systematic reviews. 13

Assessment of methodologic quality

Two reviewers, both with methodologic expertise (N.S., S.S.) and one with content expertise (S.S.), independently graded the methodologic quality of each included study using an adapted version of the Newcastle–Ottawa Scale for Cohort Studies. 14 This scale, which is intended to assess for selection and attrition bias, grades the reporting of studies on the basis of the selection, applicability and comparability of study groups; whether ascertainment of the exposure and outcome of interest was biased; and the adequacy of follow-up (ideally > 80%). 14 We deemed one item (“demonstration that outcome of interest was not present at start of study”) irrelevant and removed it from the quality assessment because our primary outcome, all-cause mortality, was unequivocal. As a result, the maximum score was 8, and the minimum score was 0. We specified a priori that a score of 7 or more indicated high methodologic quality, a score of 5 or 6 indicated moderate quality, and a score of 4 or less indicated low quality. The reviewers resolved discrepancies for each item through discussion and, when necessary, re-evaluation of the study methodology until they reached consensus.

Extraction of data

When data for a study were unclear or missing from the article or abstract, we attempted to contact the authors. For all but one of the included studies, data had been collected prospectively for the purpose of answering a specific research question. The exception 15 was a study that used prospectively collected data that had previously been entered into a hospital database. We confirmed with the authors of that study that the data had been collected accurately and consecutively for every eligible patient who had presented to two teaching hospitals, an approach that would have minimized the potential for selection bias.

Assessment of agreement

We used the κ (kappa) statistic to examine the extent of agreement between the individuals who determined study eligibility. We used the intraclass correlation coefficient to evaluate interobserver agreement in methodologic quality scores. We chose an a priori criterion of κ ≥ 0.65 to indicate adequate agreement.

Statistical analysis

We included in our primary meta-analysis those studies that adjusted the mortality results for potentially confounding variables at any follow-up time. We included in our secondary meta-analyses studies with unadjusted estimates of mortality and postoperative complications.

When frequency data were available, we calculated the relative risk (RR) and 95% confidence interval [CI] for the primary outcome (all-cause mortality) between the groups who underwent early and delayed surgery as assessed in hospital or at 30 days, at three to six months, and at one year. We grouped the data for in-hospital and 30-day follow-up on the basis of a sensitivity analysis that showed no statistical difference in mortality rates between follow-up assessments at these times. We considered adjusted estimates of mortality appropriate for the primary analysis if the authors had adjusted for at least patient age and type or severity of illness.

In the absence of frequency data, we collected the reported RR, odds ratio or hazard ratio. We converted odds ratios and hazard ratios to RRs using the methods proposed by Zhang and Yu 16 or the following formula: relative risk = 1 − e^{hazard ratio*ln(1 − P₀)}/ P₀, where P₀ is the outcome incidence in the non-exposed group.

Where appropriate, we pooled the outcome measures using the random-effects model of DerSimonian and Laird, 17 which is based on the inverse variance method. We weighted all pooled estimates by study size. We quantified heterogeneity between studies using the I² statistic, 18 which represents the percentage of total variation across trials that is due to heterogeneity rather than to chance. 19

To assess publication bias, we constructed funnel plots to examine sample size versus exposure effect across included studies. This method plots the magnitude of the exposure effect relative to the weight of the study.

Evaluation of heterogeneity

Given the potential for heterogeneity in effect sizes, we performed stratified analyses and used a statistical test of interaction 20 to evaluate the extent to which subgroup results differed from each other. We hypothesized that heterogeneity might be due to differences in the reasons for surgical delay (administrative or non-administrative), the cut-off time used to define a delay (e.g., 24 v. 72 hours), the length of follow-up (e.g., 30 days v. one year), the date of publication of the study (before the year 2000 or in the year 2000 or later) or methodologic features (low quality v. moderate or high quality).

To control for multiple testing and inflation of type I error, we defined a significant difference between subgroups as p < 0.01. We considered I² < 25% to indicate low heterogeneity and I² > 75% to indicate high heterogeneity. 21 Tests of significance for treatment effects were two-tailed, and a p value of less than 0.05 was considered significant.

Studies included

Our literature search identified 1939 potentially relevant citations: 1908 from the electronic search, 22 from hand searches of the articles that remained after initial screening (n = 98) and nine abstracts. Sixteen of these citations proved eligible for inclusion 8^,12^,15^,22^–34 (Figure 1). Two studies were possible duplicates, but we included both in our analysis because one assessed long-term mortality 24 and the other assessed short-term mortality. 25 The weighted κ for overall agreement between reviewers for the final eligibility decision was 0.85 (95% CI 0.47–1.00).

Study characteristics

The sample sizes of the included studies ranged from 65 to 3628 patients (Table 1). The cut-off times for operative delay were 24 hours, 8^,12^,15^,22^,23^,27^,28^,30^,33 48 hours, 24^,25^,29^,31^,32 72 hours 34 and five days. 26 The preoperative interval was recorded from the time of injury to surgery in five studies 12^,15^,25^,28^,29 and from the time of admission to surgery in the remaining 11 studies. Eight studies reported the reasons for surgical delay, the most common being the unavailability of an operating room and/or surgical personnel, 12^,24^,25^,27^,31^,32^,34 and investigation and stabilization of the patient’s preoperative medical condition. 22^,27^,31^,32^,34

Study quality

We judged four studies 8^,32^–34 to be of high methodologic quality, five studies 15^,22^,24^,27^,31 to be of moderate quality and the remaining seven studies 12^,23^,25^,26^,28^–30 to be of low quality (Table 1). Agreement between reviewers in the assessment of study quality was excellent (intraclass correlation coefficient 0.92, 95% CI 0.75–0.97).

Mortality

Of the 14 171 patients analyzed in all 16 studies, complete mortality data were available for 13 478. In five studies (n = 4208 patients), the researchers computed adjusted odds ratios or hazard ratios for mortality at 30 days, 32 six months 8 or one year 27^,33^,34 (721 total deaths) by means of a multivariable logistic regression model or multivariable Cox proportional hazards model. Those five studies most commonly adjusted for American Anesthetists Society score (a measure of a patient’s fitness for surgery), age and sex. On the basis of the pooled adjusted estimates, early surgery was associated with a 19% risk reduction in all-cause mortality, irrespective of the time of the outcome assessment (RR 0.81, 95% CI 0.68–0.96, p = 0.01, I² = 0%).

All 16 studies provided unadjusted estimates of mortality. The unadjusted estimates also suggested that early surgery significantly reduced the risk of one-year mortality by 45% (RR 0.55, 95% CI 0.40–0.75, p < 0.001, I² = 71%; Figure 2). Heterogeneity in the unadjusted one-year mortality rate could not be explained by the reason for surgical delay, the cut-off for delay, study quality, date of publication or length of follow-up. Funnel plots showed no evidence of publication bias.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2: Stratified analysis by time of death. Forest plot of unadjusted relative risks for the effect of early compared with delayed surgery for hip fracture on all-cause mortality assessed in hospital or at 30 days (short-term), at three to six months (medium-term) or at one year (long-term) (random-effects model based on inverse variance method). Studies used a cut-off for delay of 24 hours, except as indicated otherwise. *Study used a cut-off of 48 hours for delay. †Data based on patients who had medical illness in combination with hip fracture. ‡Study used a cut-off of 72 hours for delay. §Study used a cut-off of 5 days for delay. CI = confidence interval, RR = relative risk.

Time to surgery did not significantly affect mortality at 30 days (n = 3485; RR 0.90, 95% CI 0.71–1.13, p = 0.86, I² = 0%) or at three to six months (n = 1650 patients; RR 0.87, 95% CI 0.44–1.72, p = 0.68, I² = 87%) (Figure 2). However, the removal of a single study evaluating only medically ill patients 23 removed most of the between-study differences and resulted in a significant benefit at three to six months with earlier surgery (n = 1590 patients, RR 0.66, 95% CI 0.50–0.88, p = 0.005, I² = 7%).

Postoperative complications

Four studies 25^,29^,31^,33 reported on the number of individual postoperative complications for a total of 5377 patients. These data were not adjusted for potentially confounding surgical factors. Two of the studies 25^,33 evaluated pneumonia (n = 2793 patients, 101 total events) and demonstrated an overall unadjusted risk reduction of 41% among patients who underwent early surgery (< 24 or 48 hours) relative to those whose surgery was delayed (RR 0.59, 95% CI 0.37–0.93, p = 0.02, I² = 0%; Figure 3). In three studies, 25^,29^,33 early surgery was associated with a 52% reduction in the risk of pressure sores, on the basis of unadjusted data (n = 3023 patients, 174 total events) (RR 0.48, 95% CI 0.34–0.69, p < 0.001, I² = 0%; Figure 3).

• Download figure
• Open in new tab
• Download powerpoint

Figure 3: Forest plot of unadjusted relative risks for the effect of preoperative timing on specific postoperative complications assessed in hospital (random-effects model). Studies used a cut-off for delay of 24 hours, except as indicated otherwise. *Study used a cut-off of 48 hours for delay. CI = confidence interval, RR = risk ratio.

Two of the studies evaluated deep vein thrombosis 31^,33 (n = 4679 patients, 56 total events), and two studies examined pulmonary embolism 25^,31 (n = 2822 patients, 40 total events). Preoperative surgical delay did not significantly affect the unadjusted incidence of either of these complications (Figure 3).

Limitations

Our meta-analysis had some limitations. It is likely that our estimates of postoperative complications were subject to publication bias because we restricted data collection for that analysis to the 16 studies with data on mortality rates for patients undergoing early or delayed surgery. Our findings that early surgery may reduce the risk of pneumonia and pressure sores should therefore be interpreted with caution. In addition, we identified unexplained heterogeneity in our analysis of the unadjusted one-year mortality outcome.

The most notable limitation of this review was its restriction to observational studies, which reflects the state of current evidence. Observational studies are prone to selection, performance, attrition and detection bias. Unadjusted analyses are certainly confounded, and although we used adjusted ratios for our primary meta-analysis, the results may still be subject to confounding bias where our study may be missing other unknown or unmeasured factors potentially relevant to prognosis for a patient with hip fracture. These factors may limit the conclusions.

Conclusion

On the basis of current evidence, surgery conducted before 24–72 hours is associated with lower mortality and lower rates of certain postoperative complications among elderly patients with hip fracture. When potential confounding pre-operative factors are taken into account, this effect is not as large, but its direction is maintained. Given the challenges in interpreting observational data, there is a need for additional well-designed prospective studies or a randomized trial to offer clear insights into the effects of early surgery in this patient population.