Waiting time and mortality after hip fracture: findings from a large single-center cohort study

Abstract

Purpose

Time to surgery after hip fracture is a potentially modifiable factor that may influence mortality, but the evidence for an association is conflicting for waiting time < 48 h. The aim of this study was to investigate the association between waiting time to surgery and mortality, with waiting time categorized into 12-hour intervals.

Methods

We analysed all hip fractures operated at a large Norwegian hospital between 2015 and 2023. Waiting time from admission to surgery was categorized into 12-hour intervals. Mortality at one week, one month, three months (pre-specified primary endpoint) and one year was assessed using multivariable logistic regression, adjusting for age, sex, nursing home residency, Charlson Comorbidity Index (CCI), fracture type, calendar year of surgery, and direct-acting oral anticoagulant (DOAC) use. Model fit and discrimination were assessed for all models.

Results

The study included 4136 hip fractures. At the pre-specified primary endpoint of three months, mortality was numerically higher in patients with waiting times > 12 h, but this did not reach statistical significance for any waiting-time category (adjusted odds ratio [AOR] 1.16–1.25, all p > 0.2). Among the secondary, exploratory endpoints, the one-year timepoint showed increased mortality with increased waiting: patients operated between 12 and 24 h had significantly higher one-year mortality (AOR 1.38, 95% CI 1.10–1.72), rising to an AOR of 1.80 (95% CI 1.20–2.71) for waiting times exceeding 48 h.

Conclusions

In this cohort, waiting time beyond 12 h was associated with a higher one-year mortality, but the association was not statistically significant at the primary endpoint at three months. Given the observational design and the possibility of residual confounding, these findings should be interpreted with caution.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1007/s00068-026-03351-w.

Introduction

A hip fracture is a serious injury for an elderly person, carrying substantial risks of morbidity, loss of independence, and mortality. With ageing populations worldwide, the burden of hip fractures on healthcare systems continues to grow, making optimization of hip fracture treatment a priority. Early surgical intervention is widely regarded as a key element of effective care, and is commonly used as a treatment quality benchmark [1–2], yet the precise impact of waiting time on patient outcomes remains debated.

Several observational studies and systematic reviews have suggested that earlier surgery may reduce complications and improve survival [3–5], while others have reported no clear association between timing and mortality within the first 48 h [6–7]. Differences in study design, patient populations, and methods of adjusting for comorbidity may have contributed to these conflicting results.

Most studies investigating surgical timing in hip fracture have focused on thresholds of 24–48 h, and relatively little attention has been given to whether clinically important risk increases may occur earlier. Furthermore, large registry studies rarely include detailed information on anticoagulation use, which has been shown to be an important confounder in the relationship between surgical timing and mortality [8].

The aim of this study was to investigate the association between waiting time and mortality using 12-hour waiting time intervals in a large single-center cohort with detailed clinical data.

Methods

Design and setting

The study utilized an observational cohort design, with prospectively collected clinical data from a large orthopedic department in Oslo, Norway. This hospital manages approximately 450 hip fracture patients annually within a specialized orthogeriatric unit. Surgeries were scheduled during daytime and evening hours and were generally not performed during the night (between 23:00 and 08:00). Most of the surgeries were carried out by orthopedic surgeons in training, typically under supervision of or with the assistance of consultant orthopaedic surgeon. Since 2007, detailed information about all hip fractures has been recorded in a quality database. Mortality data were collected regularly from the national death registry.

Patients, exposures and outcomes

The study population consisted of adults aged 18 years and older who were operated for a hip fracture during the period 2015 to 2023. Patients admitted prior to 2015 were excluded due to incomplete data regarding direct-acting oral anticoagulant (DOAC) use, and those admitted after 1 January 2024 were excluded to ensure sufficient follow-up for mortality outcomes. The primary exposure was time between arrival in the emergency department and start of surgery, categorized into groups: ≤12 h, 12–24 h, 24–36 h, 36–48 h and > 48 h. Time of admission and start of surgery were collected from the electronic patient file, where the exact time of day was automatically time-stamped during the treatment of the patient. The main outcome was all-cause mortality collected from the official national death registry, assessed at one week, one month, three months and one year after admission for a hip fracture. Three-month mortality was designated as the primary endpoint, with the other timepoints considered secondary and exploratory. Adjustment covariates were age, sex, nursing home residency status, Charlson Comorbidity Index (CCI), fracture type, calendar year of surgery, and DOAC use. Body mass index (BMI) was not included as a covariate because it was missing for 18% of patients. Waiting time was defined as the interval between arrival at the hospital and start of surgery.

Statistical analysis

Although some patients had a repeated admission for a contralateral fracture, each hip fracture was analyzed as a separate case. A sensitivity analysis was performed with only the first fracture included, to evaluate if there were discrepancies in the findings. Categorical data are reported with percentages, continuous data are reported as means with standard deviations. Simple comparisons of proportions were analyzed with a chi-square test, and mean comparisons were analyzed with a Student’s t-test. Differences in baseline characteristics across waiting-time groups were compared using chi-square tests for categorical variables and one-way ANOVA for continuous variables. Kaplan-Meier survival curves were compared using the log-rank test. A test for linear trend across the ordered waiting-time categories was performed to assess the progressive gradient described in Results.

The main analysis was a multivariable logistic regression analysis with mortality as the outcome, and waiting time categorized by 12-hour intervals as exposure, with < 12 h as the reference category. Age, sex, nursing home residency status, CCI, fracture type, calendar year of surgery, and DOAC use were included as adjustment covariates. Three-month mortality was pre-specified as the primary endpoint; one-week, one-month and one-year mortality were analysed as secondary, exploratory endpoints, reflecting the increasing potential for residual confounding with longer follow-up and the number of comparisons performed. Model fit was assessed using the Hosmer-Lemeshow goodness-of-fit test and model performance was assessed using the area under the receiver operating characteristics curve. Predicted probabilities shown in Fig. 3 were calculated as adjusted (marginal) predictions from the multivariable logistic regression model, holding other covariates at their observed sample values. Trends of increasing mortality with increasing waiting time were analysed by logistic regression using waiting time as an ordinal variable.

Fig. 3.

Predicted one-year mortality by waiting-time category: (A) all patients. (B) by CCI group

The statistical significance level was set at 0.05. We did not apply corrections for multiple comparisons. All analyses were done using STATA version 17.0 (StataCorp, College Station, Tx, USA).

Prior to analysis, directed acyclic graphs were drawn using DAGITTY software [9] to clarify our understanding of the causal relationships between the studied variables (supplementary Figure. 1).

Missing data

The database is continuously quality-controlled and curated by a dedicated coordinator, leading to a high level of data completeness. Mortality, waiting time, and all covariates included in the final models were complete for all 4136 hip fractures. Body mass index was missing for 18% of patients and was therefore not included as a covariate.

Ethical issues

The study was conducted in accordance with the principles of the Declaration of Helsinki. The hospital research board and the hospital’s data protection officer has approved the quality database used for this project. Information is made available for the patients that their clinical information is recorded in the database, but written consent to participate is not required under current Norwegian regulations governing quality control registries (Health Personnel Act § 29a). Inclusion in the study posed no risk of harm to the patients.

Results

Between 2015 and 2023, 4136 hip fractures were operated, of which 676 (16.3%) were in patients using a DOAC. The mean age was 82.7 years, 70.6% were female, and 55.1% had a femoral neck fracture (Table 1). Baseline characteristics across waiting-time groups differed significantly in BMI, DOAC use, CCI and fracture type, but not in age, sex or nursing home residency (Table 1).

Table 1.

Patient characteristics in groups according to waiting time

Characteristic	< 12	12–24	24–36	36–48	48+	Total	p*
N (%)	926 (22.4%)	1704 (41.2%)	815 (19.7%)	502 (12.1%)	189 (4.6%)	4136 (100%)	–
Age, mean (SD)	82.8 (10.6)	82.5 (9.9)	82.8 (9.7)	83.3 (9.9)	82.9 (10.4)	82.7 (10.1)	0.598
Female, n (%)	673 (72.7%)	1217 (71.4%)	562 (69.0%)	342 (68.1%)	127 (67.2%)	2921 (70.6%)	0.194
BMI, mean (SD)	22.4 (4.2)	22.8 (4.4)	22.9 (4.3)	22.7 (4.3)	23.7 (5.2)	22.8 (4.4)	0.012
Nursing home, n (%)	209 (22.6%)	361 (21.2%)	174 (21.3%)	102 (20.3%)	46 (24.3%)	892 (21.6%)	0.728
DOAC user, n (%)	31 (3.3%)	142 (8.3%)	137 (16.8%)	266 (53.0%)	100 (52.9%)	676 (16.3%)	< 0.001
CCI, mean (SD)	1.5 (1.6)	1.5 (1.6)	1.7 (1.8)	1.8 (1.6)	1.9 (1.7)	1.6 (1.7)	< 0.001
CCI 0–2, n (%)	744 (80.3%)	1344 (78.9%)	609 (74.7%)	369 (73.5%)	130 (68.8%)	3196 (77.3%)	0.003
CCI 3–4, n (%)	139 (15.0%)	267 (15.7%)	142 (17.4%)	104 (20.7%)	47 (24.9%)	699 (16.9%)	0.003
CCI 5–6, n (%)	30 (3.2%)	62 (3.6%)	40 (4.9%)	19 (3.8%)	9 (4.8%)	160 (3.9%)	0.003
CCI 7+, n (%)	13 (1.4%)	31 (1.8%)	24 (2.9%)	10 (2.0%)	3 (1.6%)	81 (2.0%)	0.003
Femoral neck fracture, n (%)	470 (50.8%)	973 (57.1%)	442 (54.2%)	291 (58.0%)	103 (54.5%)	2279 (55.1%)	0.042
Trochanteric fracture, n (%)	426 (46.0%)	688 (40.4%)	354 (43.4%)	203 (40.4%)	84 (44.4%)	1755 (42.4%)	0.042
Subtrochanteric fracture, n (%)	30 (3.2%)	43 (2.5%)	19 (2.3%)	8 (1.6%)	2 (1.1%)	102 (2.5%)	0.042

* Chi-square test for categorical variable, one-way ANOVA for continuous variables

When assessing survival curves for different waiting times; the log-rank test for the difference across the five waiting-time groups (censored at 365 days) was significant (chi2(4) = 27.18, p < 0.001), with fewer deaths than expected in the < 12-hour group (170 observed vs. 219.7 expected) and more than expected in the 36–48 and 48 + hour groups (138 vs. 110.3 and 59 vs. 40.6, respectively) (Fig. 1).

Fig. 1.

Kaplan-Meier survival curves for groups of patients categorized based on waiting time. (A) Survival within the first 90 days after admission (B) 1-year survival

At the pre-specified primary endpoint at three months, the odds ratio for mortality was numerically higher with longer waiting times in adjusted multivariate analyses (Table 2; Fig. 2), but this did not reach statistical significance for any waiting-time category (AOR range 1.16–1.25, all p > 0.2; Table 2).

Table 2.

Results from the multivariable logistic regression analysis

		One-week mortality			One-month mortality			Three-month mortality*			One-year mortality
	N	n (%)	AOR (95%CI)	p	n (%)	AOR (95%CI)	p	n (%)	AOR (95%CI)	p	n (%)	AOR (95%CI)	p
< 12 h	926	17 (1.8%)	1 (ref)	–	46 (5.0%)	1 (ref)	–	97 (10.5%)	1 (ref)	–	172 (18.6%)	1 (ref)	–
12–24 h	1704	33 (1.9%)	1.09 (0.60–2.01)	0.770	99 (5.8%)	1.22 (0.83–1.78)	0.309	201 (11.8%)	1.18 (0.90–1.56)	0.238	387 (22.7%)	1.38 (1.10–1.72)	0.005
24–36 h	815	22 (2.7%)	1.41 (0.72–2.77)	0.319	59 (7.2%)	1.43 (0.94–2.19)	0.099	102 (12.5%)	1.16 (0.84–1.60)	0.369	196 (24.0%)	1.34 (1.03–1.74)	0.027
36–48 h	502	12 (2.4%)	0.93 (0.39–2.19)	0.861	48 (9.6%)	1.66 (1.01–2.72)	0.046	73 (14.5%)	1.25 (0.85–1.85)	0.254	138 (27.5%)	1.48 (1.08–2.03)	0.014
48 + h	189	6 (3.2%)	1.21 (0.42–3.44)	0.725	14 (7.4%)	1.22 (0.62–2.43)	0.564	27 (14.3%)	1.23 (0.74–2.07)	0.425	59 (31.2%)	1.80 (1.20–2.71)	0.004
Model diagnostics
Hosmer-Lemeshow p		0.22			0.16			0.22			0.02
AUC (c-statistic)		0.822			0.798			0.778			0.781

AOR: adjusted Odds Ratio, adjusted for age, sex, nursing home status, Charlson Comorbidity Index, fracture type, calendar year of surgery, and DOAC use. *Pre-specified primary endpoint. Model diagnostics: Hosmer-Lemeshow goodness-of-fit test (higher p = no evidence of lack of fit), AUC = area under the ROC curve

Fig. 2.

Adjusted odds ratios and 95% confidence intervals from the multivariable logistic regression analysis

Among the secondary, exploratory endpoints, this pattern reached statistical significance at one year: patients operated between 12 and 24 h had significantly higher one-year mortality compared to patients operated within 12 h (AOR 1.38, 95% CI 1.10–1.72). One-year mortality risk continued to increase with longer delays, reaching an AOR of 1.80 (95% CI 1.20–2.71) for waiting times exceeding 48 h. At the remaining secondary timepoints — one week and one month — point estimates were directionally similar but confidence intervals were wide and with one exception not statistically significant. The trend of increasing mortality with increasing waiting time, was tested using waiting time as an ordinal variable, and was found to be statistically significant at one year (p = 0.004), but not at the earlier timepoints.

We investigated the effect of increased waiting time for different CCI groups. Because the proportion of fatalities was small in the different subgroups at the early time-points, this analysis was performed only for one-year mortality. Adjusted one-year mortality increased with waiting time within every CCI stratum (Fig. 3B), broadly parallel increases rather than a clearly widening gap. None of the twelve individual waiting-time / CCI-group interaction terms in the underlying logistic model reached statistical significance, and their point estimates (odds ratios 0.57–1.42) showed no consistent direction, providing no support for effect modification by comorbidity burden.

A sensitivity analysis including only the first fracture in patients with two contralateral fractures found no substantial differences in effect estimates (Supplementary Table 1).

Discussion

In this cohort study of hip fracture patients, we observed higher mortality among patients whose surgery was delayed beyond 12 h after admission, an association that was statistically significant at one year but not at the pre-specified primary endpoint at three months. Given the observational design and the pattern of statistical significance, these findings are far from conclusive.

Waiting time before surgery for hip fracture is a topic of significant interest, as it is a major modifiable factor in the care for a vulnerable group. A large study from Ontario, Canada, found that a delay longer than 24 h was associated with a higher risk-adjusted likelihood of 30-day mortality [4]. Similarly, a systematic review by Klestil et al. showed that early surgery (within 24–48 h) is associated with reduced mortality and fewer complications [10]. Beaupre et al. [11] found that the effect of prolonged waiting time was more pronounced in patients with higher age, and also on those with male gender and increased comorbidity. Warren et al. found increased 30-day survival if patients were operated within 12 h, although their analysis compared the < 12-hour group with all other patients, including those waiting more than 72 h, where the mean waiting time was more than 5 days [3]. Several other large meta-analyses also found an association between waiting time and increased mortality [12–14]. However, a large registry study from the Norwegian Hip Fracture Registry found no significant differences in mortality between patients operated in the similar 12-hour categories up to 48 h after admission, but mortality was increased in patients waiting more than 48 h [7].

The question of whether very rapid surgery is beneficial for the survival of hip fracture patients was also investigated by the HIP ATTACK trial [15]. In this multi-center, randomized trial with 2970 patients included, no reduction in mortality or major complications was found with accelerated surgery (target: <6 h within diagnosis) compared with standard care (median 24 h). As a randomized trial, HIP ATTACK provides higher-quality evidence than observational cohort studies such as the present one. The discrepancy between its findings and our observation of higher mortality beyond 12 h therefore reinforces the possibility that residual confounding or other biases may partly explain the association observed in our data. However, randomized trials operate under controlled conditions and may not reflect the longer delays and system-level variation seen in routine practice, limiting external validity.

In our study, the association between waiting time and mortality reached statistical significance at one year, while estimates at one week, one month and three months — the endpoint we pre-specified as primary — were directionally similar but did not reach significance for most of the waiting-time categories. If a delay beyond 12 h had a direct, mechanistic effect on early mortality, one might expect this to become apparent earlier rather than emerge only after a year of follow-up. Because the association reached statistical significance and the largest effect estimate at this longest and least specific follow-up window, the one-year finding must be interpreted with caution.

Two explanations may be considered. First, with relatively few early deaths, the study may simply be underpowered to detect a moderate early effect, and a true effect might only become statistically detectable once a larger number of events has occurred, i.e. after one year. Another more likely explanation is that the one-year estimate is also the one most susceptible to residual and unmeasured confounding, also highlighted by the poorer model fit in the Hosmer-Lemeshow test.

If the association between waiting time and mortality observed here is at least partly causal, several biologically plausible pathways could contribute. Prolonged preoperative delay may increase exposure to hospital-acquired complications such as pneumonia, urinary tract infection, delirium, venous thromboembolism and acute kidney injury, which can contribute to earlier death in frail older patients. Extended delay may also increase pain, systemic inflammatory responses and catabolism, potentially exhausting cardiorespiratory reserves. Mobilisation is delayed, potentially leading to a less optimal rehabilitation period.

Several limitations should be considered when interpreting these findings. First and most importantly, this is an observational study, and the potential for residual confounding from unmeasured variables such as acute physiological derangement at presentation, detailed frailty indices, pre-fracture functional status and the precise clinical reasons for delay cannot be excluded. Antithrombotic therapy illustrates this limitation, as this may both influence surgical timing and identify patients with greater underlying morbidity and vulnerability, but its inclusion as a single covariate cannot fully separate these two roles. In a recent study, antithrombotic therapy was consistently associated with delayed surgery, but the associations with mortality were substantially attenuated after adjustment for patient characteristics [16], illustrating how this variable may act as a marker of vulnerability rather than as a confounder fully captured by covariate adjustment. As detailed in a paper by Sheehan et al. [17], the relationship between surgical timing and mortality after hip fracture is complex, with many patient and system factors influencing both the surgical timing and the outcome. The observed associations may therefore not be causal. Second, as discussed above, the association reached clear statistical significance mainly at one year, our least specific and most confounding-prone endpoint, this pattern argues for caution rather than a straightforward causal reading of the one-year result. Third, we did not apply a correction for multiple testing across the five waiting-time categories and four timepoints examined, which increases the risk of type I error for the secondary, exploratory endpoints; only the three-month timepoint was pre-specified as primary. Fourth, the single-center setting and local operating practices, including a scheduled absence of overnight surgery, may limit the external validity of the results. Finally, small numbers in the most delayed categories produced imprecise estimates with wide confidence intervals, and the discrepancy between our findings and the null result of the HIP ATTACK randomized trial, discussed above, further underscores that the association reported here should not be interpreted as proof of a causal effect of surgical delay on mortality.

Implications and further research

The results support systematic efforts to minimize avoidable waiting time and to prioritize early surgery. They support guidelines such as the current Norwegian national recommendation stating that hip fractures should be operated within 24 h after admission [1], and the recommendation from the UK´s Falls and Fragility Fracture Audit Programme, that states that hip fractures should be operated on the day of admission or the day after [2]. Future research should prioritize multicenter studies and registry enhancements that prospectively capture anticoagulation details, standardized frailty and comorbidity measures, the recording of explicit reasons for waiting time and comprehensive perioperative complication data. A given waiting time may reflect very different underlying processes, ranging from avoidable delay due to limited operating-room capacity to clinically necessary delay for medical stabilization or management of anticoagulation, and distinguishing between these categories may be more clinically informative than a single time threshold. This distinction should be a priority for future data collection.

Conclusion

In this cohort study, prolonged waiting time after admission was not significantly associated with higher mortality during the first three months. For one year mortality, however, waiting more than 12 h after admission was associated with increased mortality, and the mortality increased with increasing waiting time. The observational design cannot exclude residual confounding, and the results may be attributable to bias.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

MS designed the study, performed the analysis, and wrote the manuscript. US contributed to design, analyses and writing. JS contributed to the design of the statistical analysis and writing of the manuscript. MM organized the data collection. SL has been central in both funding procurement, analysis and writing. All authors have reviewed and approved the final manuscript before submission.

Funding

Open access funding provided by Diakonhjemmet Hospital. The corresponding author currently has a research grant from the South-Eastern Norway Regional Health Authority.

Data availability

The study is based on data from the Diakonhjemmet hip fracture database. The possibilities of data sharing are restricted due to privacy regulations, but it may be considered upon a reasonable request.

Declarations

Ethical approval

According to Norwegian legislation, this study is classified as quality control research, and approval of the research has been done by the relevant internal bodies and the data protection officer of the hospital.

Ethics

The study was conducted in accordance with the principles of the Declaration of Helsinki, and inclusion in the project was not associated with any conceivable harm.

Consent to participate

The database is classified as a quality control database, and written consent is not required under current Norwegian regulations governing quality registries.

Competing interests

The authors declare no competing interests.