The use of electronic medical records for hip fracture risk stratification.

摘要

Background and Objectives Hip fracture risk stratification currently requires a provider assessment and often involves a radiology measurement of bone mineral density. It is unknown whether electronically available patient information is useful in predicting hip fracture. The goals of this project are, first to determine whether an Elders Risk Assessment (ERA) Index, an electronic medical records (EMR) derived frailty index, can passively predict incident hip fractures, second to use EMR to create hip fracture risk assessment index (HFRAI), an index that is exclusive for hip fracture risk stratification, and third to compare HFRAI to the world health organization fracture risk assessment tool (FRAX). Subjects and Methods A retrospective cohort study was conducted in a community-based cohort of 12650 subjects 60 years and older. For the first study: An ERA score was computed for each subject at (January 1, 2005). The outcome was incidents of hip fracture in the subsequent two years. We divided the cohort into 5 groups, with the lowest 15th percentile ERA scores forming Group A, 15%-49% Group B, 50%-75% Group C, 75%-90% Group D and the top 10% Group E. With group A as a reference group, we used logistic regression to compute odds ratios (OR) for groups B, C, D and E. We also created a Receiver Operating Curve (ROC). For the second study: The total cohort was randomly divided into a development cohort (N=8,387) and a validation cohort (N=4,263). Using the development cohort¡¦s EMR, risk factors were evaluated for significance in univariate and multivariate logistic regression. The outcome was incident hip fracture in the subsequent four years. Risk factors were assigned a score based on their regression coefficient and a total risk score HFRAI was created. We used EMR of the validation cohort to evaluate HFRAI and create an ROC. For the third study: We randomly selected 1700 (850 with a known FNBMD and 850 without known FNBMD). We applied FRAX algorithm to obtain 10-year probability of hip fracture for each subjects as of 01/01/2005. Using HFRAI scores and FRAX output with the primary outcome of incident hip fracture in the subsequent four years, we created three pairs of ROCs, one pair for each of total 1700 subjects group, and the two subgroups of subjects with and without known FNBMD. We used Z-test statistics to compare areas under the curve for each pair. Results First study: Two hundred sixty-five subjects (2.1%) sustained at least one hip fracture in 2005 and 2006. ORs (95% CIs) for groups B, C, D and E were 1.6 (0.7-3.4), 4.5 (2.2-9.4), 6.9 (3.3-4.3), and 18.4 (8.9-37.9), respectively. The area under ROC was 74.5%. Second study: 288 (3.43%) subjects in the development cohort sustained a new hip fracture. The final model included age ,,d75 (odd ratio (OR) 10, 95%CI 4.4, 23), age 70-74 (OR 7.7, 95%CI 3.3, 18.1), age 65-70 (OR 3.8, 95%CI 1.6, 9.4), white race (OR 2.5, 95%CI 1.2, 5.5), female gender (OR 1.4, 95%CI 1.1, 1.9), prior hip fracture (OR 5.2, 95%CI 3.7, 7.3), liver cirrhosis (OR 2.7, 95%CI 1, 7.5), malnutrition (OR 2.5, 95%CI 1.5, 4.2), peripheral arterial disease (OR 1.6, 95%CI 1.2, 2.3), Parkinson¡¦s disease (OR 1.6, 95%CI 1.2, 2.2), hyperparathyroidism (OR 1.9, 95%CI 1.1, 3.3), prior nursing home stay (OR 1.7, 95%CI 1.2, 2.3), and prior hospitalization (OR 1.4, 95%CI 1.1, 1.8). Area under receiver-operating curve (AUC) was 0.8, SE 0.017. Subjects in the highest 10% of the risk group had (OR 8.8, 95%CI 6.3, 12.3); for incident hip fracture versus the rest of the group. Third study: 62 of 1700 study subjects (3.6%) sustained a hip fracture between 01/01/2005 and 12/31/2008 (34 patients with known FNBMD and 28 patients without known FNBMD). In the total group AUC for our created index was 0.75, which was no different than AUC for FRAX of 0.71 (p=0.19). Similarly, there was no significant difference in AUCs between our index and FRAX for each of the two subgroups. Conclusion Electronically obtained, readily available patient information can be utilized for hip fracture risk stratification, without performing further patient evaluation or FNBMD testing. We demonstrated that an EMR derived frailty index, ERA, can predict hip fracture. We also created and internally validated a hip fracture risk assessment index (HFRAI) that only uses EMR data and found HFRAI comparable to FRAX in a cohort of community dwelling older adults.