Impact of timing of surgery in elderly hip fracture patients: a systematic review and meta-analysis

We aimed to assess the impact of timing of surgery in elderly patients with acute hip fracture on morbidity and mortality. We systematically searched MEDLINE, the Cochrane Library, Embase, PubMed, and trial registries from 01/1997 to 05/2017, as well as reference lists of relevant reviews, archives of orthopaedic conferences, and contacted experts. Eligible studies had to be randomised controlled trials (RCTs) or prospective cohort studies, including patients 60 years or older with acute hip fracture. Two authors independently assessed study eligibility, abstracted data, and critically appraised study quality. We conducted meta-analyses using the generic inverse variance model. We included 28 prospective observational studies reporting data of 31,242 patients. Patients operated on within 48 hours had a 20% lower risk of dying within 12 months (risk ratio (RR) 0.80, 95% confidence interval (CI) 0.66–0.97). No statistical significant different mortality risk was observed when comparing patients operated on within or after 24 hours (RR 0.82, 95% CI 0.67–1.01). Adjusted data demonstrated fewer complications (8% vs. 17%) in patients who had early surgery, and increasing risk for pressure ulcers with increased time of delay in another study. Early hip surgery within 48 hours was associated with lower mortality risk and fewer perioperative complications.

Factors that influence prognosis of elderly patients after hip fracture are age, gender, comorbidities, anticoagulation therapy, and general physical health status at the time of injury 24 . Furthermore, timing of surgery is thought to play an important role regarding survival. Although international clinical practice guidelines recommend surgical treatment of acute hip fracture within 24 to 48 hours after admission [25][26][27] , these recommendations are still discussed controversially [28][29][30] . Some researchers argue that early surgery can lead to an increased risk of perioperative complications, including pneumonia, deep venous thrombosis, bleeding, pulmonary embolism, urinary tract infections, and decubital ulcerations because clinicians do not have enough time to optimise patients' medical conditions preoperatively [29][30][31] .
The most recent systematic review on this topic was published in 2010 32 . Since then, many well-conducted studies have been published. To provide a comprehensive overview, it is necessary to systematically review the currently available evidence on the impact of timing of surgery in elderly patients with acute hip fracture. In contrast to former reviews that focused exclusively on mortality, we additionally aimed to assess other patient-relevant outcomes, such as perioperative complications, functional capacity, and quality of life. We also explored whether timing of surgery has different effects in different subgroups, e.g., in patients on anticoagulation treatment or patients with poor physical status.
Our systematic review aimed to answer the following questions: (1) In patients aged 60 years or older with an acute hip fracture, what is the impact of timing of surgery on beneficial and harmful outcomes such as mortality, functional capacity, quality of life, and perioperative complications? (2) Do beneficial or harmful treatment effects of timing of surgery vary by subgroups based on patient characteristics (age, sex), physical status (e.g., ASA Physical Status System), and common medical treatments (e.g., anticoagulation treatment)?

Methods
To answer our research questions, we conducted a systematic review that has been registered with the International Prospective Register of Systematic Reviews (PROSPERO), registration number: CRD42017058216 33 . The study protocol has been published previously 34 . We will summarise the most important methodological steps in the sections below.

Search Strategy and Criteria. An experienced information specialist searched MEDLINE (Ovid),
PubMed (non-MEDLINE content), Embase.com, the Cochrane Library (Wiley), for the period of January 1997 to May 2017, using keywords and medical subject headings for hip fracture surgery, adult patients, and timing factors. To ensure finding all relevant studies on this topic a broad range of synonyms where used for the search (see Appendix 1 for the search strategy). In addition, we searched the World Health Organization (WHO) International Clinical Trials Registry Platform (ICTRP) and ClinicalTrials.gov, as well as reference lists of relevant publications, websites and conference proceedings of orthopaedic and traumatological societies (see Appendix 2).

Inclusion and Exclusion.
Inclusion and exclusion criteria were predetermined in the published protocol 35 .
Eligible study designs were randomised controlled trials (RCTs), non-randomised controlled trials, and prospective controlled cohort studies. The populations of interest were adults aged 60 years or older undergoing surgery for acute intra-and extracapsular hip fracture. We also included studies where only a small proportion (<5%) of patients were younger than 60 years. Studies were included only if they compared early and delayed surgery for hip fractures. The primary outcome was all-cause mortality. Secondary outcomes of interest were perioperative complications, functional capacity, and quality of life. Detailed eligibility criteria are presented in Table 1.

Assessment of Study Quality and Certainty of Evidence.
We used the Newcastle-Ottawa-Scale (NOS) to judge the risk of bias in included cohort studies 36 . Two authors independently assessed the risk of selection bias, comparability of groups, adequacy of outcome measurement, and reporting. We resolved disagreements by consensus or involvement of a third review author. In addition, we assessed the certainty of evidence (CoE) across studies for important outcomes following recommendations of the Grading of Recommendations, Assessment, Development and Evaluation (GRADE) working group 37 . Experts in the field of orthopaedics and traumatology ranked outcomes regarding clinical and patients' relevance in a modified two-staged Delphi process. They agreed on mortality, quality of life, perioperative complications, and function/mobility as the most important outcomes. For these outcomes, we graded the certainty of evidence and classified it as "high, " "moderate, " "low, " or "very low. " High certainty means we are very confident that the true effect is close to the effect estimate. On the contrary, if the certainty is very low, we assume that the true effect is likely to be significantly different from the effect estimate 37 .
Data Collection and Abstraction. Two review authors independently reviewed abstracts and full-text articles in two consecutive steps. Disagreements were resolved by consensus or discussion with a third author. Two team members independently extracted relevant information on study design, methods, patient characteristics, intervention, control, and outcomes from included studies. In case information about relevant outcomes or study characteristics was missing or unclear, we contacted study authors.
Meta-analysis Methodology. We used the generic inverse variance method to combine effects of individual observational studies that were adjusted for potential confounders and were rated as low or moderate risk of bias in meta-analyses. We pooled data only if at least three studies used comparable cut-offs for "early" and "delayed" surgery and reported the same outcome. In case the studies reported hazard ratios (HR) or odds ratios (OR), we converted them into risk ratio (RR) using the following formulas for HR: RR = 1 − e (HR* ln (1 − P 0 )) /P 0 32 , and for OR: RR = OR/((1 − P 0 ) + (P 0 * OR)) 38 ; P 0 means the event rate in the control group. For one study 39 , we were not able to calculate P 0 because no crude numbers of events were reported, so we used the mean P 0 from the other included studies to convert OR into RR. We added observational studies with unadjusted results, irrespective of their risk of bias judgment to meta-analyses for sensitivity analyses.
To assess statistical heterogeneity in effects between studies, we calculated the chi-squared statistic and the I 2 statistic (the proportion of variation in study estimates attributable to heterogeneity rather than due to chance) 40,41 . Due to the limited number of studies included in meta-analyses, no funnel plots could be used to assess publication bias. We used RevMan Version 5.3 42 for all statistical analyses.
For outcomes for which no meta-analyses were possible, we summarise data narratively. If several studies reported the same outcome but meta-analyses were not possible because of high clinical heterogeneity or because the study was rated high risk of bias, we graphically display results in forest plots without pooled summary estimates.
Because data were not sufficient to conduct subgroup analyses, we summarise these results narratively.

Results
Study characteristics. We included 28 prospective cohort studies 13 Cut-off 48 hours. Based on a meta-analysis of adjusted data from four studies 13,39,62,68 the absolute risk of dying within 12 months was 21% in patients who had surgery after 48 hours and 17% in patients who had surgery within 48 hours resulting in a 20% smaller long-term mortality risk in patients operated on within 48 hours (RR 0.80, 95% CI 0.66-0.97, 2,396 patients, see Fig. 2). We graded the CoE for this outcome as low. We also conducted sensitivity analyses by adding unadjusted data on mortality from the remaining studies to the meta-analysis,

Population
• Studies including at least 95% adults aged 60 years or older who underwent surgery for acute hip fracture (intra-or extracapsular)  Fig. 7). Figure 2 summarises results of meta-analyses on long-term mortality and corresponding sensitivity analyses. Data were insufficient to conduct meta-analyses for other cut-offs (6, 12, 18, 36, 72 hours) of timing of surgery. However, to illustrate the results on mortality of all studies, we present forest plots for each cut-off in Appendix 3 (48 hours: see Six studies reported unadjusted data for perioperative complications 29,48,54,55,58,64 . Figure 3 presents unadjusted effect estimates of individual studies. While a cut-off of 6 hours did not show significantly different rates of complications, patients who had surgery within 24 or 48 hours suffered from complications less frequently than those with late surgery. One study on 744 patients used three different cut-offs (24, 36, 48 hours) for "delayed surgery" and presented adjusted data for pressure ulcers.  62 . In studies reporting unadjusted data the risk for developing pressure ulcers, pneumonia, urinary tract infections, or thromboembolic events was either smaller for patients who had early surgery or similar between both groups; it was not higher in any study for patients who had early surgery (see Fig. 4). CoE for perioperative complications was very low.   Functional capacity. Measuring of mobility and functional capacity was different among eight studies 29,44,48,56,57,61,63,66 and data are summarised in Table 3. Patients who had early surgery had similar or slightly better functional capacity compared to those operated on later (CoE: very low).

Impact of timing of surgery in subgroups.
Due to insufficient data, we were not able to conduct subgroup analyses to assess different effects of timing of surgery between age groups, sex, patients' physical status, and anticoagulation. However, six studies assessed the effects of timing of surgery in different subgroups 29,47,54,57,58,61 . Below we present results narratively.
Age. In two studies, timing of surgery (before or after 24 hours) showed no significant difference in mortality rates in different age groups. Yonezawa 61 . Again, the low number of events makes chance findings inevitable. Another study divided patients into two groups, either fit or unfit for immediate surgery, depending on their physical status. In the group of patients considered fit for surgery, no statistically significant difference between early (within 24 hours) and delayed surgery was observed regarding 30-day mortality (85/982; 9% vs. 85/1166; 7%; p = 0.510) 47 . In the group of patients with acute medical comorbidities, there was no significant relationship   In patients with lower POSSUM scores no difference in mortality between early (within 48 hours) and delayed surgery was reported (POSSUM score 37-40: early: 8% vs. delayed: 11%, p = 0.500; POSSUM score ≤ 36: early: 24% vs. delayed: 50%, p = 0.310).
Pioli et al. hypothesised that timing of surgery is more important for frail elderly patients than for older people without functional impairment. Therefore, they divided patients into three groups according to their IADL (Instrumental Activities of Daily Living) score. One-year mortality in group 1 (dependent) and group 2 (intermediate level) relatively increased by 14% and 21%, respectively, per day of surgical delay (HR 1.14; 95% CI 1.06-1.22, p < 0.001 and HR 1.21; 95% CI 1.09-1.34, p < 0.001), but not in group 3 (high independence; HR 1.05; 95% CI 0.79-1.41, p = 0.706) 57 .
In a prospective cohort study including 1,206 patients, those with abnormal clinical findings or the need for further preoperative evaluation were excluded to form a restricted cohort of medically fit patients. In this group, early surgery within 24 hours had no association with functional outcomes or mortality, but was associated with reduced major postoperative complications (p = 0.041) 29 .
Anticoagulation treatment. In most of the studies, anticoagulants were more common in the delayed group and frequently caused surgical delay 13,39,48,51 . However, we did not identify any study reporting on differences between early and delayed surgery in patients with and without anticoagulation treatment.

Discussion
To the best of our knowledge, this is the first systematic review critically assessing all relevant prospective studies on this topic since 2010. We identified 20 new studies that had not been considered in the previous reviews 32, 70,71 . Our findings agree with previous systematic reviews. Simunovic et al. showed that early surgery (within 24 to 72 hours) can reduce the risk of all-cause mortality in patients aged 60 or older by 19% (risk ratio (RR) 0.81, 95% confidence interval (CI) 0.68-0.96) 32 . Early surgery was also associated with a reduction of pressure ulcers and postoperative pneumonia (RR 0.48, 95% CI 0.34-0.69) 32 . Another systematic review including prospective and retrospective observational studies also demonstrated that a delay in surgery beyond 48 hours was associated with an increased 1-year-mortality and 30-day mortality risk (odds ratio (OR) 1.32, 95% CI 1.21-1.43; 30-day mortality: OR 1.41, 95% CI 1.29-1.54) 70 .   In contrast to other systematic reviews we looked at the effect of different cut-offs for "early" and "delayed" surgery separately and found that early surgery within 48 hours was associated with decreased long-term mortality in elderly patients after hip fractures. Single studies using other cut-offs (6,12,18,24 or 36 hours) did not demonstrate significant differences in mortality between early and delayed surgery. However, these studies were probably underpowered and it is important to note that no study demonstrated a beneficial effect of delayed surgery on mortality.
Although findings of this review strengthen existing guidelines recommending surgery within 48 hours, in clinical practice, delay of surgery of hip fractures is quite common. In situations where patients need medical optimisation due to poor health status or long-term medication 72 , delays cannot be avoided. However, the reasons for delayed surgery are also often limited capacity of operating rooms and personnel, or weekend and holiday administration 32,58,73,74 . Cha et al. showed that hospital factors are accountable for three-fourths of the surgical delays 74 . In the interest of high quality care, organisational and structural improvements, such as better availability of operating rooms and staff, are necessary to enable early surgery. There is also general agreement that rapidly correctable comorbidities such as anaemia, hypovolemia, electrolyte imbalance, and correctable cardiac arrhythmias should not delay the operation 27 .
Only six of the included studies reported the effects of time to surgery in our predefined subgroups. In healthy, independent patients, delayed surgery was not as problematic as in patients with comorbidities. In most of these studies, the event rate was very small. Hence, the results could be chance findings. Moreover, the studies presented only unadjusted data. It should be emphasised that conclusions based on this data must be drawn carefully. Nevertheless, if availability of staff and operation room is limited, comorbid patients could be prioritised and have early surgery, presupposing that they do not have clear contraindications for surgery.
Our study has some limitations. We graded the certainty of evidence for all outcomes low or very low, which means that our confidence in the findings is limited. One reason for the low certainty of evidence is that we only identified prospective cohort studies but no RCTs. Results of observational studies must be interpreted with caution since confounding could distort the findings. It is possible that non-organisational reasons for delay of surgery such as need for medical optimisation also increased the risk of dying, independently, or in addition to   timing of surgery. To minimise the distortion through confounding we included for our main analysis only data from adjusted analyses where at least the most important confounders such as age, gender, ASA score, fracture type and comorbidities had been considered. However, due to lack of randomisation, confounding cannot be completely eliminated.
The studies identified used different cut-offs to define early and delayed surgery. We combined only data from studies using very similar cut-offs. This allowed us to include only a small number of the included studies into meta-analyses. However, presenting the evidence for different cut-offs separately is relevant to inform clinical practice about the optimal timing of surgery.
No study conducted subgroup analysis with tests for interaction. However, some analysed the effect of timing of surgery in separate strata, allowing us to draw some conclusions about different effects in subgroups. Moreover, often the number of events was very small, making chance findings very likely. The results on subgroups therefore have to be interpreted with caution.
Despite our comprehensive search, it is possible that not all studies conducted on this topic have been detected (e.g., studies published in languages other than English or German). Publication bias cannot be ruled out, and we were not able to assess potential publication bias with a funnel plot. However, we contacted experts in the field, searched trial registries, and ultimately found 20 new studies that have not been included in former systematic reviews.
To overcome the limitation of observational studies, RCTs on this topic are needed. Although experts often argue, that this is unethical and not possible to implement, a RCT on timing of surgery in hip fracture patients is on the way. The HIP-ATTACK trial (HIP fracture Accelerated surgical TreaTment And Care tracK) will compare the effect of accelerated surgery and standard surgical care on perioperative complications and mortality 75 . A total of 1,200 patients older than 45 with low-energy hip fracture will be included in the study. The results of this trial will inform clinical practice and for the first time control adequately for known and unknown confounders.

Conclusion
In elderly patients sustaining hip fracture, early surgery is associated with reduced mortality and perioperative complications. Patients operated on within 48 hours had a 20% lower 1-year mortality.  However, timing of surgery for patients with hip fractures remains a challenge, as it requires multidisciplinary coordination between different occupational groups and the availability of appropriate surgical capacity with competent staff and proper equipment. No study demonstrated a survival benefit with delayed surgery. Future studies should investigate the effect of early surgery in subgroups of patients (e.g. patients with greater co-morbidities or anticoagulation treatment) and include data on patient-relevant outcomes, such as quality of life measurements. Furthermore, randomised controlled trials are needed to rule out potential confounding.

Data Availability
The datasets generated during the study are available from the corresponding author on reasonable request.