Separating the signal from the noise: how mortality rate associated with hip fracture changes over time after accounting for population level mortality rates

It is important to measure how survival following hip fracture changes over time because it is an important patient-centred outcome and a reflection of the care we offer our patients. Thirty-day mortality outcomes after hip fracture are published on a rolling basis by the National Hip Fracture Database (NHFD) [1]. Whereas case-mix adjusted mortality rates are useful for identifying hospital outliers from the national average, they are not well suited to identifying temporal changes in mortality rates attributable to improving standards of care. Case-mix adjusted mortality rates do not adjust for variations in the mortality risk of the general population, such as those associated with the COVID-19 pandemic and, importantly, the progressive long-term trend of declining mortality risk over time. Furthermore, 30 days is too brief a period to capture the impact of many important interventions designed to improve survival [2, 3].

The purpose of this study was to apply an alternative analysis to existing hip fracture mortality data to distinguish improvements in mortality due to better care from changes due to fluctuating population mortality rates. All patients meeting the inclusion criteria for the NHFD between 2012 and 2023 were identified within a single London hospital. The survival status, including date of death if applicable, was established for each patient by searching the electronic patient record and the National Spine database. Patients were not included if they did not have surgery or their survival status could not be confirmed. Patients aged < 65 y and > 95 y were not studied due to limited numbers. Identified patients were segregated into four trienniums (2012–2014, 2015–2017, 2018–2020 and 2021–2023) according to the date of fracture. Each triennium was stratified by sex and 5-year age categories enabling the observed deaths at 1 month and 12 months to be calculated for each stratum. The expected deaths for each stratum were calculated using the mortality rates published in the Office for National Statistics lifetables [4]. These are published in trienniums matching those used in this study. Excess mortality rates for each triennium are presented as standardised mortality ratios (SMR), calculated using the indirect method with 95%CI. The SMR represents the ratio of observed to expected deaths, adjusted for age, sex and year of fracture.

The basic demographic data were consistent over the four trienniums (Table 1). The overall crude mortality rate was 5.08% at 1 month and 21.12% at 12 months.

Table 1. Patient characteristics for each triennium post hip fracture surgery. Values are median (IQR [range]) or 95%CI.

	2012–2014	2015–2017	2018–2020	2021–2023
	n = 348	n = 484	n = 539	n = 649
Female:male ratio	71:29	67:33	71:29	67:33
Age; y	84 (76–88 [65–94])	83 (76–88 [65–94])	82 (75–88 [65–94])	82 (76–87 [65–94])
Standardised mortality rate post-surgery
1 month	7.92 (4.70–12.52)	6.47 (4.00–9.88)	6.54 (4.10–9.91)	8.30 (5.71–11.65)
12 months	2.46 (1.90–3.12)	2.23 (1.79–2.75)	2.60 (2.13–3.15)	2.77 (2.31–3.38)

This study shows that the excess mortality rate associated with hip fracture in our hospital has not improved over the period 2012–2023. Johnston et al. applied a similar methodology to a Scottish cohort in 1998–2005 and reported an absolute mortality rate of 30.7% at 1 year, compared with 21.1% in this study [2]. However, the background mortality risk experienced by the general population in the period 1998–2005 was substantially higher than this more recent study. Accounting for this, Johnston et al. reported the SMR at 1 year to be 1.89, which is lower than any of the trienniums reported here. This comparison shows how misleading absolute mortality rates can be. The SMR describes the change in mortality risk attributable to hip fracture more accurately between the two cohorts and paints a very different picture.

This is a single-centre study limited by relatively small numbers, requiring 3-year cohorts and 5-year age stratification. Ideally SMRs would be calculated for each individual year and patients would be stratified in single year age ranges. This method assumes the national population mortality risk describes background mortality risk of our local population exposed to hip fracture accurately. This is unlikely for two reasons. First, there is regional variation in mortality risk and our London population is likely to have lower background mortality risk than the national average. Second, patients with hip fracture present with greater comorbidities than their age and sex matched population average. These limitations should be recognised but barring significant demographic changes over the 12-year study period the impact will be consistent over time. Solving these limitations with propensity score matching would not have the repeatability advantage that calculating SMRs offers.

This study shows a straightforward method for tracking changes in excess mortality following hip fracture surgery. It has the advantage of enabling comparison of mortality outcomes between populations of different baseline mortality risk. Therefore, temporal comparisons can be made without attributing incorrectly the well-established trend of improving population level mortality risk to improvements in clinical care. Further efforts should focus on reproducing this method in a multicentre study. Larger numbers across multiple sites will mitigate the limitations mentioned above. It would be valuable if these data were produced annually, allowing appropriate monitoring of mortality outcomes across time.