A commonly used assessment scale to predict 1-year mortality in the general population is the CCI. (14) However, it is unclear whether the CCIs predicted 1-year mortality aligns with the actual clinical outcomes of patients with hip fractures. Based on the high 30-day mortality rate within the hip fracture population (5, 7-9, 23), it is also helpful to know whether a CCI is related to a higher 30-day mortality. The aim of our study was to validate the original CCI and adjusted CCI from 2011 as risk prediction models for 30-day and 1-year mortality after hip fracture surgery.
The mean of the original CCI in this cohort was 5.1 (SD±2.0) and 4.6 (SD±1.9) for the adjusted CCI. A previous study of a hip fracture population reported a similar CCI value of 5 (SD±2). (24) The observed 30-day and 1-year mortality rates in this study were 8.6% and 25.9%, respectively. This is in accordance with previously reported 30-day mortality rates between 6.4% to 13.3% and 1-year mortality rates between 23.2% and 33.0%. (5-9)
Variables of the CCI
After univariable analysis, an age above 71 years was associated with a higher 30-day mortality rate, and an age above 61 years was associated with a higher 1-year mortality. Increasing age is a well-established risk factor for both 30-day and 1-year mortality after hip fracture surgery. (5, 7-9, 25) It seems justifiable that the CCI increases with age.
Based on our study results and a recently published meta-analysis, the medical history of dementia had a substantial influence on mortality after hip fracture. (26) It seems appropriate that the weight of dementia in the adjusted CCI has been increased from 1 to 2. The 30-day and 1-year mortality rates were significantly higher in patients with diabetes, which is in line with previous literature. (27) Nevertheless, the weight of diabetes was decreased after adjustment. Congestive heart failure has been described in the literature as a risk factor for in-hospital and 1-year mortality. (28-30), which is in line with our univariable analysis results. In our cohort, MI was associated with higher 30-day and 1-year mortality. MI is not a well-described variable in the literature in relation to mortality in hip fracture patients. Nevertheless, Guzon-Illescas et al. also reported a relation between a higher mortality and MI in medical history. (30) The fact that MI is no longer included in the adjusted CCI is debatable for hip fracture patients. MI can be an ambiguous variable since a history of MI does not provide any information on the remaining heart function, which may be more relevant. A medical history of cerebrovascular disease was associated with increased 1-year mortality. Previous studies have reported no correlation between prior cerebrovascular accidents and in-hospital or 1-year mortality after hip fracture surgery. (28, 30, 31) By varying results, it is discussable that cerebrovascular accidents are no longer included in the adjusted CCI for hip fracture patients.
Moderate or severe renal disease was significantly associated with a higher 1-year mortality rate. There were no significant differences in mortality among patients with mild, moderate or severe liver disease in our cohort. This is probably due to the small number of patients in our cohort. Patients with local tumors, metastatic tumors and leukemia had significantly higher 1-year mortality. Lymphoma and acquired immunodeficiency syndrome (AIDS) were not significantly associated with 30-day or 1-year mortality after univariable analyses.
Chronic pulmonary disease, connective tissue disease, hemiplegia, peripheral vascular disease and peptic ulcer disease were not associated with increased mortality in our cohort. These factors have not been described in the literature as risk factors for mortality in hip fracture patients. A higher CCI is well described in the literature as a risk factor for both 30-day and 1-year mortality. (8, 11, 32, 33) This finding is in line with our univariable analysis results for both the original and adjusted CCIs.
The CCI and mortality rates
The distributions of the original and adjusted CCIs of all patients and patients who died within 30 days and 1 year after hip fracture surgery are visualized in histograms (Figure 3). On average, the original CCI was approximately 1 point higher in the 30-day mortality and 1-year mortality groups than in patients who did no died. The adjusted CCI showed a slightly larger difference between the entire cohort for 30-day mortality of approximately 2 points but also appeared to be 1 point greater for 1-year mortality than in patients who did no died. Ideally, by predicting mortality, the differences between the mortality and nonmortality groups are more clearly reflected in the histogram.
The percentages of observed mortality for each score of the original CCI and the adjusted CCI are shown in Table 2. The mortality rate continued to increase to approximately 9 points on the CCI. After a CCI of 9 points, a decrease in the mortality rate can be perceived, possibly because of the small number of patients included in these categories. The observed 1-year mortality rates of the original CCI scores are described in the article by M. Charlson: CCI 0 = 12%; CCI 1-2 = 26%; CCI 3-4 = 52%; and CCI 5 = 85%. (14) In our cohort, these percentages were 0%, 3%, 13% and 35%, respectively, for the original CCI and 0%, 4%, 15% and 38%, respectively, for the adjusted CCI (Table 2). The 1-year mortality percentages from 1984 were not comparable to the 1-year mortality percentage from this cohort for either the original CCI or the adjusted CCI.
Performance and validation of the CCI
The performance statistics of the original and adjusted CCIs are shown in Figure 4. The AUCs were 0.6736 and 0.696 for 30-day mortality for the original and adjusted CCIs, respectively. The AUCs for 1-year mortality were 0.705 and 0. 717 for the original and adjusted CCIs, respectively. Tang et al. reported an AUC of 0.653 for in-hospital mortality. (34) The adjusted CCI was previously validated by Haugan et al., who reported an AUC of 0.726 for 30-day mortality and 0.751 for 1-year mortality. (21) However, Haugan et al. did not report mortality rates, studied each variable of the CCI, generated calibration plots and did not validate the original CCI. Karres et al validated 6 prediction models for 30-day mortality after hip fracture surgery and found an AUC of 0.72 for the original CCI. (22)
Other prediction models for 30-day and 1-year mortality for hip fracture patients have recently been developed. (13, 15) The NHFS was validated by Sun et al. and reached an AUC of 0.791 for 30-day mortality. (35) The BHFS was internally validated and had AUCs of 0.71 and 0.75 for 30-day and 1-year mortality, respectively. (15) Other prediction models do allow direct calculation of the mortality risk for a specific patient; hence, a certain CCI score does not directly translate into a predicted 30-day or 1-year mortality rate.
The calibration curves for 30-day mortality for both the original and adjusted CCIs did not show good consistency between the CCI and observed 30-day mortality. The CCI did not differ between the patients with approximately 20% predicted mortality or higher, and the same mortality was observed for all those patients. Therefore, there is overprediction from approximately 20% of the predicted mortality. The calibration curve for 1-year mortality showed a better fit for the original and adjusted CCI, but the curve also strongly deflects from the 40% observed mortality for both the original and adjusted CCI.
Strengths
In this study, we investigated a large cohort of hip fracture patients with detailed descriptions of baseline and perioperative factors in a prospective hip fracture database. The cohorts are representative of the target population and have limited missing data (including limited missing data for the CCI variables), increasing the external validity of the findings. The data on the primary outcome measures, 30-day mortality and 1-year mortality, were complete.
A complete overview of the performance of the original and adjusted CCI for both 30-day and 1-year mortality prediction after hip fracture surgery was established. The sample size and particularly the number of events of the test cohort included more than 100 patients, and almost all baseline factors (including the variables of the CCI) were complete, indicating that adequate validation could be conducted. (36)We performed extensive external validation, including both statistical and graphical assessments of the discrimination and calibration, which made the results more reliable than those of previous studies. This is the first study in which the use of the CCI is not recommended for the prediction of 30-day and 1-year mortality in hip fracture patients, as better prediction models are currently available.
Limitations
This was an observational cohort study based on patients’ medical charts, meaning that potentially unreported data were not included in our analysis. However, the comprehensive database encompasses all consecutive patients, and due to careful status research and follow-up, the amount of missing data was very limited. The CCI may be lower in this cohort than in the whole population because of missing data on diabetes with end-organ failure and the severity of liver disease. Nevertheless, this difference should not be large due to the relatively low incidence of these diseases in a broader population. Additionally, we excluded the most fragile patients who were treated nonoperatively because we aimed to investigate risk factors for 30-day mortality after hip fracture surgery.