Attributable Mortality of Hip Fracture in Older Patients: A Retrospective Observational Study

Hip fracture (HF) in older patients is associated with a high six-month mortality rate. Several clinical conditions may affect outcome, including baseline characteristics, co-existing acute illnesses, perioperative factors, and postoperative complications. Our primary objective was to estimate the respective effect of these four domains on six-month mortality after HF. A retrospective observational study using a monocentric cohort of older patients was conducted. All patients ≥ 70 years old admitted to the emergency department for HF and hospitalized in our perioperative geriatric care unit from June 2009 to September 2018 were included. Among 1015 included patients, five (0.5%) were lost to follow-up, and 1010 were retained in the final analysis (mean age 86 ± 6 years). The six-month mortality rate was 14.8%. The six-month attributable mortality estimates were as follows: baseline characteristics (including age, gender, comorbidities, autonomy, type of fracture): 62.4%; co-existing acute illnesses (including acute events present before surgery that could result from the fracture or cause it): 0% (not significantly associated with six-month mortality); perioperative factors (including blood transfusion and delayed surgery): 12.3%; severe postoperative complications: 11.9%. Baseline characteristics explained less than two-thirds of the six-month mortality after HF. Optimizing patients care by improving management of perioperative factors and thus decreasing postoperative complications, could reduce by a maximum of one quarter of the six-month mortality rate after HF.


Introduction
Background/rationale 2 Explain the scientific background and rationale for the investigation being reported Page 3 Objectives 3 State specific objectives, including any prespecified hypotheses Page 3

Study design 4
Present key elements of study design early in the paper Page 4 Setting 5 Describe the setting, locations, and relevant dates, including periods of recruitment, exposure, follow-up, and data collection Page 4 Participants 6 (a) Give the eligibility criteria, and the sources and methods of selection of participants. Describe methods of follow-up Page 4 (b) For matched studies, give matching criteria and number of exposed and unexposed Variables 7 Clearly define all outcomes, exposures, predictors, potential confounders, and effect modifiers. Give diagnostic criteria, if applicable Page 4-6 Data sources/ measurement 8* For each variable of interest, give sources of data and details of methods of assessment (measurement). Describe comparability of assessment methods if there is more than one group Page 4-6 Bias 9 Describe any efforts to address potential sources of bias Page 6-7 Study size 10 Explain how the study size was arrived at Page 6 Quantitative variables 11 Explain how quantitative variables were handled in the analyses. If applicable, describe which groupings were chosen and why Page 6-7 Statistical methods 12 (a) Describe all statistical methods, including those used to control for confounding Page 6-7 (b) Describe any methods used to examine subgroups and interactions (c) Explain how missing data were addressed (d) If applicable, explain how loss to follow-up was addressed (e) Describe any sensitivity analyses Results

Participants
13* (a) Report numbers of individuals at each stage of study-eg numbers potentially eligible, examined for eligibility, confirmed eligible, included in the study, completing followup, and analysed   Management strategy in UPOG (previously detailed)[1] focused on early mobilization with the aim of chair-sitting and walking within 24 and 48 hours after arrival respectively, pain management using acetaminophen and morphine, the provision of air-filled mattresses for patients with pressure sores or a high risk of pressure sores as evaluated by the Braden scale (8), swallowing disorders detected using a systematic medical survey, detection of stool impaction and urinary retention using bedside ultrasound, correction of anemia with transfusion of packed red blood cells (usually when the hemoglobin level was <8 g.L -1 ), [2] detection of delirium using the Confusion Assessment Method, [3] and malnutrition detection and management in conjunction with a nutritionist. We recorded the preoperative hemoglobin level, and anemia was defined according to World Health Organization criteria.
[4] We calculated preoperative serum creatinine and estimated creatinine clearance using the Cockcroft-Gault formula. Chronic renal failure was defined as a creatinine clearance  30 ml.min -1 .
Post-operative complications included postoperative delirium, need for physical restraints, stool impaction, urinary retention requiring drainage, pressure sore, infection, aspiration related to swallowing disorders, thromboembolic events, need for blood transfusion, cardiac insufficiency, myocardial infarction, acute atrial fibrillation, stroke, surgical complications, and admission into an intensive care unit (ICU).  if follow-up could not be obtained.

Primary endpoint (response variable)
The primary endpoint is 6-month mortality status. Patients are classified using two experts, when disagreement occurs a third expert is used. The kappa score is calculated.

Statistical methods
Patient characteristics will be described overall and according to 6-month mortality status.
Quantitative variables will be described by their mean, standard deviation, median, interquartile range, minimum, maximum value and number of missing data. Qualitative variables will be described by frequency, percentage and number of missing data. This descriptive analysis will be performed in the overall sample and in two subgroups: alive and dead at 6 months.
The model of prediction of 6-months mortality will be constructed in the sample with no missing values for all candidate explanatory variables.
In the spirit of parsimony and to keep the final model as simple as possible, we will 1) separately select the most important variables in each of the four domains, 2) and then fit the final model with all the variables selected in the previous step.

a. Separately select the most important variables in each domain
Separately for each domain, all continuous variables will be categorized, either through clinically relevant thresholds from the literature, or using ROC curve to determine the best threshold (maximization of the Youden index). Univariate comparison between survivors and dead patients will be performed, using Student t test, Mann-Whitney test, Chi square test, or Fisher's exact test, as appropriate. All variables will be included in a multivariate logistic model with 6-month mortality as the explained variable. OR and their corresponding 95% confidence interval (95%CI) will be provided for each variable. Discrimination and calibration of multivariable the model will be assessed using cstatistics and Hosmer-Lemeshow test. This model will be used to determine the most important variables of the corresponding domain.

b. Fit the final model
A final model will be constructed using the most important variables of each domain selected in the previous step. No further selection of variables will be performed. Again, discrimination and calibration will be evaluated, and OR will be provided with their 95%CI. Moreover, Averaged Attributable Fractions (AAF) will be computed for each variable and each domain.[1-3]

c. Sensitivity analyses
The final model and the AAF associated with each domain depend on the way the most important variables of each domain are selected in the first step. Several methods will be used: -Selection of the significant variables of each domain-specific multivariate logistic model. Two thresholds will be used: p < 0.05 (primary analysis), and p < 0.1.
-Selection of the 3 variables with the greater AAF in each domain-specific multivariate logistic model.
The first method of selection (p < 0.05) and the corresponding final model and AAFs will constitute our primary analysis. The other methods will constitute the sensitivity analyses and will be reported as supplementary analysis.
All statistical analysis will be performed using R (version 3.5.1 or higher) software.