In this study, patients hospitalized for hip fractures at a university hospital from May 2016 to December 2021 were collected. Relevant medical record data is extracted from the electronic medical record system, and the patient’s prognosis is followed up to determine the patient’s specific survival period. This data was accessed on December 26, 2022 for research purposes. Inclusion criteria: patients who were hospitalized for hip fracture. Exclusion criteria: 1. age < 60 years; 2. pathological fractures; 2. multiple injuries; 3. patients with comorbid cancer, acute cardiac and cerebral diseases, and those who had been bedridden for a long time before the fracture; 4. partial absence of data. Similar inclusion and exclusion criteria were used in studies related to perioperative blood transfusion and postoperative pneumonia in hip fracture patients during this period at our study center (13, 14).

Data was extracted from three primary categories: patients’ baseline characteristics, commonly reported chronic conditions in geriatric populations, and key laboratory test results obtained within 24 h of admission. Additionally, we specifically included whether the patient underwent surgery. (1) General characteristics of the patients: gender (Female), age, fracture side (Left), fracture time, and fracture type (femoral neck fracture, FNF). (2) Common chronic diseases in the elderly: hypertension (HBP), coronary heart disease (CHD), diabetes mellitus (DM), cerebral infarction (CI), and chronic bronchitis (CB). (3) Laboratory test results: white blood cell count (WBC), neutrophil ratio (N), red blood cell count (RBC), hemoglobin (HB), platelet count (PLT), glucose (GLU), alanine transaminase (ALT), aspartate transaminase (AST), total bilirubin (STB), direct bilirubin (DBIL), indirect bilirubin (IBIL), albumin (ALB), globulin (GLOB), blood urea nitrogen (BUN), creatinine (Cr), potassium ion (K⁺), sodium ion (Na⁺), calcium ion (Ca²⁺), prothrombin time (PT), international normalized ratio (INR), fibrinogen (FIB), activated partial thromboplastin time (APTT), thromboplastin time (TT), prothrombin activity (PTA). For laboratory test results, the normal range may vary from hospital to hospital depending on the type of test. Therefore, to make the model generalizable, we converted it into a dichotomous variable by its exceeding the upper limit or being below the lower limit in combination with the clinical. Data extraction and follow-up were performed independently by two researchers, and a third independent researcher verified the accuracy of the data.

The study was approved by the hospital’s ethics review committee (ethics number: KYXM-202212-017, approved on December 15, 2022). An informed consent waiver was obtained because the study was retrospective and we withheld patients’ personal information in the analysis. All procedures complied with the 1964 Declaration of Helsinki and its subsequent revisions.

The mean standard deviation is used to represent continuous variables that follow a normal distribution, the median (interquartile range) is used to represent continuous variables that do not follow a normal distribution, and percentages are used to represent categorical variables. The univariate analysis of the data between the two groups was done with the “CBCgrps” package. Besides, we also analyzed the difference between the different types of hip fractures in order to compare the differences between the two groups of intertrochanteric fractures and femoral neck fractures.

All of the analysis’s patients were randomized 70:30 into the training set (70%) and validation set (30%) groups. To identify the key variables associated with mortality in elderly people one year after hip fracture, we first performed correlation tests for all variables. The correlation coefficient takes values in [−1, 1], the larger its absolute value, the stronger the correlation, greater than 0.4 indicates that there is a significant correlation, and the results are plotted as a hotspot plot. If there is significant multicollinearity between the variables, we will use the least absolute shrinkage and selection operator (LASSO) (15, 16) technique to screen the variables; otherwise, univariate logistic regression will be followed by multifactor logistic regression analysis. A linear model was fitted to the screened variables and backward stepwise regression was used to analyze the extent to which all the included variables contributed to the model. P < 0.05 was defined as significant, the variables with insignificant contributions were excluded, and the remaining ones were used as the final incorporated variables to construct the model. And based on the results, a nomogram was drawn and a dynamic web calculator was constructed. The receiver operating characteristic (ROC) (17, 18) curves were used to assess the predictive performance of the nomogram. The higher the area under the curve (AUC) of the ROC, the better the model’s performance. Sampling was performed 1,000 times using Bootstrap to assess the stability and generalization performance of the models. The impact of each variable on the overall model performance was assessed by removing each variable one by one for sensitivity analysis, constructing sub-models and comparing the AUC of each model. The calibration curve was used to indicate the agreement between the predicted and actual probabilities of the analyzed model (19, 20), and the Hosmer–Lemeshow test was used to evaluate the nomogram’s goodness of fit (21). In addition, decision curve analysis (DCA) and clinical impact curves (CIC) were used to assess the clinical utility of the nomogram in decision making (22). R (version 4.1.3) software was used to implement all statistical analyses, model development, and validation in this study.

Based on the inclusion and exclusion criteria, 619 patients were ultimately included in the study. The screening and entire analysis process are shown in Figure 1. There were 318 femoral neck fractures and 301 intertrochanteric fractures in the included patients, of which 10.7% (n = 34) and 21.3% (n = 64) were treated conservatively. 136 (21.97%) patients died within 1 year of the fracture. Table 1 summarizes the characteristics of all enrolled patients and divides them into two groups to examine the differences according to whether they died one year after the fracture. Compared to the survival group, the death group exhibited significant differences in several variables. For instance, the proportion of females was 54% in the death group versus 66% in the survival group; the median age was 84 years in the death group compared to 78 years in the survival group; the time from fracture to hospital admission was longer in the death group (P = 0.043); 38% of patients in the death group had CHD versus 15% in the survival group; 35% had CI compared to 24% in the survival group; 20% of patients in the death group had B versus 9% in the survival group; 43% of patients in the death group sustained femoral neck fractures compared to 54% in the survival group; and only 57% of the death group underwent surgical treatment, in contrast to 97% in the survival group. Furthermore, the differences between intertrochanteric fractures and femoral neck fractures are presented in Supplementary Table 1. All patients were randomized into the training set (n = 435, 70%) and the validation set (n = 184, 30%), and baseline information for both groups is shown in Table 2.

We extracted 35 variables from each patient. The results of the correlation test between variables showed significant multicollinearity among several variables (23) (Figure 2). Based on the results of the correlation test, we chose LASSO regression to perform variable screening. The results showed that when the lambda value was chosen as lambda.1se (0.05750), a total of eight variables with non-zero coefficients were screened (Figure 3). In order to further simplify the model, we build generalized linear models using the filtered variables and assess the relative contributions of each component. After excluding 2 variables with insignificant contributions, the remaining 6 variables (age, CHD, surgery, HB, AST, and BUN) were used to construct the final prediction model (Table 3) and visualized as a nomogram (Figure 4a). In addition, we designed an online web calculator to assess the 1-year risk of death after hip fracture in older adults based on the results (Figure 4b)¹.

We plotted ROC curves based on the nomogram model. The AUC values of the nomogram model were 0.83 (95% CI: 0.78–0.88) in the training set (Figure 5a) and 0.81 (95% CI: 0.73–0.89) in the validation set (Figure 5b), respectively. Similarly, the results from 1000 bootstrap resamples indicate that the nomogram exhibits good stability and generalizability, as detailed in Supplementary Table 2. After sequentially excluding Age, CHD, Surgery, HB, AST, and BUN, the model’s AUC values were 0.82, 0.83, 0.79, 0.82, 0.82, and 0.82, respectively. It shows that the model has strong discriminative ability in both the training and validation sets. The calibration curves of this prediction model show good agreement between predictions and observations in the training set (Figure 5c) and the validation set (Figure 5d), with the Hosmer–Lemeshow test showing non-significant p-values of 0.55 and 0.12, respectively. The DCA curves show that for both the training set (Figure 5e) and the validation set (Figure 5f) if the threshold probability is in the range of 0.1 to 0.9, the prediction model achieves a higher net return than the “all intervention” or “no intervention” strategy. The CIC results showed that the number of high-risk patients (number of deaths predicted using the nomogram) and the number of incident high-risk patients (number of true-positive deaths) were highly matched, with threshold probabilities above 0.62 in the training set (Figure 5g) and 0.65 in the validation set (Figure 5h), respectively.