Meta-analysis and metaregression of risk factors associated with mortality in hip fracture patients during the COVID-19 pandemic

Incidence of hip fractures has remained unchanged during the pandemic with overlapping vulnerabilities observed in patients with hip fractures and those infected with COVID-19. We aimed to investigate the independent impact of COVID-19 infection on the mortality of these patients. Healthcare databases were systematically searched over 2-weeks from 1st–14th November 2020 to identify eligible studies assessing the impact of COVID-19 on hip fracture patients. Meta-analysis of proportion was performed to obtain pooled values of prevalence, incidence and case fatality rate of hip fracture patients with COVID-19 infection. 30-day mortality, excess mortality and all-cause mortality were analysed using a mixed-effects model. 22 studies reporting 4015 patients were identified out of which 2651 (66%) were assessed during the pandemic. An excess mortality of 10% was seen for hip fractures treated during the pandemic (OR 2.00, p = 0.007), in comparison to the pre-pandemic controls (5%). Estimated mortality of COVID-19 positive hip fracture patients was four-fold (RR 4.59, p < 0.0001) and 30-day mortality was 38.0% (HR 4.73, p < 0.0001). The case fatality rate for COVID-19 positive patients was 34.74%. Between-study heterogeneity for the pooled analysis was minimal (I2 = 0.00) whereas, random effects metaregression identified subgroup heterogeneity for male gender (p < 0.001), diabetes (p = 0.002), dementia (p = 0.001) and extracapsular fractures (p = 0.01) increased risk of mortality in COVID-19 positive patients.

Information sources and search strategy. The literature search strategy was developed in collaboration with a senior information specialist and was performed over a 2-week period, 1st-14th November 2020. The Healthcare Databases Advanced Search (HDAS) interface developed by the National Institute for Heath and Care Excellence (NICE) was used to conduct a comprehensive search of the EMBASE, MEDLINE and EMCARE databases as well as the Cochrane Register of Studies (CRS) (CENTRAL) databases. A combination of controlled vocabulary and free text terms was used without any language constraints. The search strategy is presented in Supplementary Appendix 1.
Selection process and data collection. Titles and abstracts were initially screened by two independent authors (FJR and DMR) and full-text articles based on eligibility and inclusion criteria were reviewed. Data was extracted by two review authors (FJR and DMR) using a spreadsheet. Author name, year of publication, type and design of study, study period, sample size, patient characteristics, COVID-19 status, comorbidities using the Charlson Comorbidity Index (CCI), social status, cognitive status and frailty scores using the Nottingham Hip Fracture Score (NHFS) or the Clinical Frailty Score (CFS). Outcomes in terms of mortality, post-operative complications and length of stay were also recorded.
The included studies in this review were performed over different time periods during this pandemic. Due to varying infection rates worldwide, we evaluated its impact as a possible contributor to inter-study heterogeneity. Through published data on the prevalence of COVID-19 infection and associated hospital occupancy 20,21 , we obtained estimates of the average 14-day COVID-19 positive cases (COVID-19 prevalence) and number of patients in-hospital with COVID-19 for each country included in the study period. Data items. The following data was collected from the included studies: • Article (Author, year, journal of publication) • Study design (Sample size, type of study) • Study population and demographics (Age, gender, comorbidities) • COVID-19 prevalence • Trauma (Patterns, fracture type, management, type of fixation) • Hospital quality measures (Length of stay, time to surgery, rehabilitation) • Outcomes Outcomes and prioritization. Our outcome measures included all-cause mortality due to concurrence COVID-19 infection, excess mortality when comparing outcomes during pandemic and pre-pandemic controls, in-patient mortality and 30-day mortality.

Risk of bias in individual studies.
The critical appraisal for methodological quality was performed by two review authors independently (FJR and DMR) and discrepancies were resolved by a third author (RB). The Newcastle-Ottawa scale (NOS) for non-randomized studies was used, with a range of 0-9 22,23 . Synthesis and statistical analysis. A descriptive synthesis summarised study characteristic, patient demographics and reported outcomes. Where substantial heterogeneity in study design and population demographics occurred, a narrative review was used to analyse this data. Meta-analysis using a mixed effects model www.nature.com/scientificreports/ was only performed when no evidence of substantial design and study characteristic heterogeneity was found. We calculated excess mortality during the COVID-19 pandemic to evaluate its true impact on hip fracture mortality irrespective of the direct deaths caused by COVID-19 infection. Our aim was to capture COVID-19 deaths that were not correctly diagnosed or missed, in addition to indirect deaths from other causes attributable to the overall crisis. Moreover, mortality in all hip fractures testing positive for COVID-19 was compared to nonpositive COVID-19 mortality and defined as all-cause mortality. Pooled dichotomous outcomes were analysed to obtain estimates of odds ratio (OR) or risk ratio (RR) and associated 95% confidence intervals (CI) 24 . For late outcomes (e.g., 30-day mortality) a time-to-event data meta-analysis was performed using the inverse variance method to obtain summary hazard ratios (HR) with 95% CI. Based on a paper by Tierney et al. 25 a mixture of direct (e.g. results from COX regression models or reported HR's and 95% CI) or indirect methods (e.g. reported log-rank test p-value with events to patient ratios or estimates from published survival curves) was applied to calculate the individual study HR and standard error (SE) for outcome measures. A random-effects meta-regression was performed to assess potential sources of heterogeneity for studies reporting COVID-19 hip fracture case fatality and mortality rates 26 . Estimates for declared COVID-19 prevalence, hospital occupancy by country and known risk factors were incorporated into the meta-regression model. Sensitivity analyses. Sensitivity analysis was performed to evaluate the robustness of the observed outcomes and compare studies rated as low or moderate risk of bias and assess against potential confounders in all studies reporting adjusted and unadjusted results. A Newcastle-Ottawa Score (NOS) of 5 or more has been shown to be moderate or good quality rating of papers, hence this cut-off was used for sensitivity analysis 27 . Heterogeneity of included studies. Inter-study heterogeneity was assessed using calculated X 2 and I 2 statistic, whereby a X 2 p-value < 0.05 and I 2 < 50% suggested low heterogeneity. Publication bias was assessed using funnel plots for outcomes reported by 22 studies and Egger's test assessed for small study effects 22,23 . All analyses were performed on STATA 16 (Stata-corp, College station, Texas, USA).

Results
Systematic review search results. Through the search, 146 studies were obtained from Embase, 65 from Emcare and 151 from Medline. After removal of duplicates, 84 studies were screened, out of which 62 full-text articles were assessed for eligibility. 40 were excluded as they did not report on mortality specifically, leaving a total 22 studies to be included, as shown by the PRISMA diagram in Fig. 1. Studies reporting COVID-19 positive patients only, as well as those comparing them to a COVID-19 non-positive cohort and a pre-pandemic cohort were included. Grey literature such as conference abstracts, non-peer reviewed articles or letters were not included due to concern over the quality of rapid research work produced during the pandemic. There is a need for robust peer-reviews and strict measures to ensure that integrity of evidence synthesis is maintained 27 . Study characteristics and methodological assessment (Newcastle-Ottawa Scale). No randomized-controlled trials were found during the search period. All studies included in this review were case-control studies and observational cohort studies, with seven of prospective [28][29][30][31][32][33][34] and fifteen of retrospective 7,13,[35][36][37][38][39][40][41][42][43][44][45][46] design. The Newcastle-Ottawa scale (NOS) was used to assess methodological study quality, with a range of 1-9 (poor to good), as shown in Supplementary Table. All included studies in this review were susceptible to selection and timing bias due to the non-standardised COVID-19 testing protocols and inconsistent study periods whereby included patients may not be truly representative of true prevalence. The study characteristics are summarised in Table 1.

Risk of bias.
The distributions of calculated effect sizes were plotted against the precision of each study (standard error) on funnel plots which were found to be symmetrical, as shown in Fig. 2a-d. Furthermore, there was no evidence of publication bias due to small study effects based on Egger's test for all 3 meta-analysis models, i.e., case fatality rate, all-cause mortality and 30-day mortality, with p = 0.21, 0.17 and 0.36 respectively.
Deceased patients' characteristics in meta-analyses. The pandemic witnessed a total of 331 deceased patients amongst the included studies. Out of these, 180 were infected with COVID-19, and 151 were tested as COVID-19 negative. The remaining 9 were either not tested or considered as suspected. The mean age range was 84-93. 5 Non-positive COVID-19 patients during the pandemic vs pre-COVID-19 control. Seven studies [31][32][33][34]43,44,46 reported the number of non-COVID positive deaths during the pandemic period, which was meta-analysed against the mortality in the pre-COVID-19 control cohort. An odds ratio (OR) of 0.97 (95% CI 0.67-1.39), p = 0.08 was observed, as shown in Fig. 4b.

Meta-regression. Impact of COVID-19 infection and hospital occupancy. Meta-regression was performed
to evaluate the impact of COVID-19 prevalence on pooled mortality estimates as shown in Table 4. Significance was seen between all-cause mortality, COVID-19 prevalence (p = 0.01) and hospital occupancy (p = 0.05) suggesting a positive association, as shown in Fig. 7a,b. No correlation or association was seen between 30-day mortality, prevalence and hospital occupancy due to COVID-19 infection.

Discussion
This meta-analysis provides an in-depth review of the impact COVID-19 has had on the mortality of patients with hip fractures. We identified independent predictors of poor outcomes in hip fracture patients testing positive for COVID-19 and have demonstrated a four-fold increased risk of mortality in this cohort following admission (RR 4.59) and a 30-day mortality of 38% (HR 4.73). Moreover, the overall case fatality rate for COVID-19 positive hip fracture patients was 34.74% which is substantially higher than reported case fatality rates for patients with COVID-19 infection, ranging from 3.5 to 20.8% with increasing age 48 . Despite various containment protocols and mitigation strategies adopted by countries during the pandemic, the incidence of hip fractures has remained unchanged 7,31 . Independent of COVID-19 infection, patients with hip fractures have a reported 30-day mortality of 7.5-10% 49 . Hospitalization of these patients may subject them to additional risk given their vulnerability to COVID-19 in an overburdened healthcare system, potentially resulting in suboptimal healthcare provision 39,40,42 . This is apparent as we observed a mortality of 10% in all hip fractures, with a two-fold increase (OR 2.00, p = 0.007) in excess mortality. Whilst this rise may theoretically be due to the indirect impact of the pandemic irrespective of the patients' COVID-19 positive status, our analysis has shown no overall increased risk when comparing non-positive COVID-19 patients to the pre-pandemic cohort (OR 0.97, p = 0.08). This suggests that having a COVID-19 infection may independently impact the excess mortality observed.  www.nature.com/scientificreports/ With uncertainty around the novel variants and lack of definitive treatment, the need to enforce measures to reduce the spread of this disease is essential to mitigate mortality. Through our meta-regression, we observed that prevalence of COVID-19 disease (number of positive cases/100,000) and hospital occupancy due to COVID-19 directly affect the all-cause mortality of patients with hip fractures (p = 0.01 and 0.05 respectively). This highlights the concern that a non-linear rise in mortality risk may be seen if tight infection control measures are not implemented, due to healthcare systems being overwhelmed by critically unwell patients. We observed a high pooled case fatality rate (38.9%) from studies performed in the USA, specifically New York, with a sixfold increased risk of mortality in COVID-19 positive hip fractures. This may be explained by the region being the epicentre of the pandemic and reflected by the high prevalence of COVID-19 infection and associated hospital occupancy further supporting the impact this pandemic has had through straining of healthcare resources 50 .
The pre-morbid status of patients has been shown to independently contribute to adverse outcomes in patients with an isolated hip fracture or COVID-19 infection. The vulnerability of these groups subjects them to a far greater risk of poor outcomes, as highlighted in the included studies 7,35,36,39,41,47 . Known predictors such as advanced age, male gender, frailty, multiple comorbidities, dementia and cognitive impairment, ASA grade (American Society of Anesthesiologists), baseline ambulation and residential status are well established risk factors of mortality in hip fracture patients 51 . Many of these risk factors overlap with known predictors of COVID-19 mortality from recent studies 52 . Our findings have reflected this through a positive association seen between hip fracture mortality, male gender (p < 0.001), diabetes (p = 0.002) and dementia (p = 0.001), which to our knowledge are novel findings.
Recent studies have suggested that the mortality of COVID-19 patients may be due to the virally driven cytokine storm response 53,54 which subjects patients to an increased risk of thromboembolic events and could exacerbate the hypoxaemia seen in COVID-19 related acute respiratory distress syndrome (ARDS) 55,56 . A similar cytokine mediated inflammatory response has been studied in patients with hip fractures, where the cytokine kinetics curves were higher in patients with worsened outcomes 57 . This supports the "two-hit theory" proposed by various studies 40,42,58 whereby the pro-inflammatory state induced by the stress of injury, coupled with a "second-hit" resulting from surgical insult may exacerbate inflammation in acutely ill COVID-19 patients. Whilst this might skew the decision towards conservative management, our study showed that patients who had surgical repair still had a more favourable outcome. Another factor weighing into this decision is the time to surgery. Whilst the recommended timeframe for surgery is within 36 h, six studies 7,32,35,[39][40][41]47 reported a time to surgery > 36 h. Seven studies 7,28,31,35,36,42,47 reported a higher mortality with extracapsular fractures amongst patients with COVID-19 which is supported by our pooled estimate (RR 1.78, p = 0.012). This is in line with established evidence of the poorer outcomes observed for such injuries as patients susceptible to extracapsular fractures are often older, with hip osteoarthritis requiring osteosynthesis 59 . This leads to a slower recovery, longer length of hospital stay with an increased risk of nosocomial infections, and prolonged surgical procedures in unstable injuries 51 . Additionally, four studies 7,35,42,47 reported an increased mortality with intramedullary fixation, as shown in our results (RR 1.33, p = 0.04) in patients with COVID-19 infection. The obvious difference between this type of implant and other forms of extracapsular fixation is the instrumentation of the femoral canal which is known to be associated with increased mortality due to increased intramedullary pressure, embolic showers and fat extravasation and may be catastrophic to COVID-19 patients, representing a "second-hit" postulated by Lebrun et al. and Egol et al. 39,42,60,61 .
The change in theatre organisation, with the appropriation of additional steps to accommodate aerosol generating procedure (AGPs), has resulted in an increase in operative delay over the COVID-19 period, as shown by Narang et al. and Segarra et al. 31,32 . The former 31 has however observed a faster time to surgery for COVID-19 infected patients, possibly due to a conscious decision to expedite surgery in an attempt to improve outcomes for these patients. The benefits of early intervention (within 36 h) is well-known in the literature 62 . To overcome the hurdles imposed by COVID-19, Malik et al. 43 implemented an multidisciplinary (MDT) approach to facilitate decision making, resulting in a reduced COVID-19 mean time to surgery compared to pre-COVID-19 era (21.8 h vs 28.2 h) as well as a shortened time from admission to orthogeriatric assessment.  www.nature.com/scientificreports/ This apparent discrepancy between prevalence and mortality in our analysis might be due to missed opportunities to identify and prioritise management of COVID-19 positive hip fracture patients who are in the highest risk cohort. Infectivity is dynamic and being in hospital increased the risk of viral transmission. Whilst the aim should be to avoid prolonged inpatient stay, the health-burden of COVID-19 may not necessarily allow this. Supporting this, five authors 7,32,39,42,46 observed an increased mean LOS. However, by carefully risk stratifying patients and deciding the management plan, Malik et al. showed a statistically reduced LOS of patients during the pandemic, thus minimising risks involved with transmission (8.6 vs 16.3).
There are several limitations to our meta-analysis. The studies included ranged from case series to multicentre studies with varying testing protocols which may affect the true representativeness of the study population limiting the conclusions drawn in this review. Only studies in English were included which may introduce further selection bias. Moreover, the majority of studies were performed in Europe, and this imbalance of sources increases the possibility of publication bias. In several studies, clinical parameters were not clearly defined in addition to varying follow-up times. Moreover, we observed heterogeneity in the range of symptoms, interventions and outcomes reported amongst the studies included due to a lack of objective measurements. Despite these limitations, our study is the first to quantify the independent impact of COVID-19 infection on hip fracture mortality. Furthermore, we have identified modifiable variables through our analysis which can impact outcomes for vulnerable patients potentially enabling a better surgical risk stratification. Hence, there is a requirement for more robust evidence through larger samples and more reliable testing methods to further establish the true impact of COVID-19 on hip fracture outcomes.

Conclusion
Our study has shown an increased overall and 30-day mortality of hip fracture patients treated during the COVID-19 pandemic with concomitant COVID-19 infection being an independent risk factor of mortality. We highlight the impact prevalence and hospital occupancy has had on mortality as surrogate markers of overburdened healthcare systems. We believe the vulnerability of hip fracture patients increases with peak incidence of COVID-19. Thus, their care must be prioritised during this crisis through means of a comprehensive care pathway. Furthermore, we identified modifiable predictors of poor outcomes in COVID-19 positive hip fractures, such as male gender, diabetes, dementia and intramedullary fixation devices, enabling clinicians to mitigate risk and aid decision-making during this pandemic. www.nature.com/scientificreports/