OBJECTIVE: To evaluate the accuracy of using ICD-9 codes to identify nonunions (NU) and malunions (MU) among adults with a prior fracture code and to explore case-finding algorithms. STUDY DESIGN: Medical chart review of potential NU (N=300) and MU (N=288) cases. True NU cases had evidence of NU and no evidence of MU in the chart (and vice versa for MUs) or were confirmed by the study clinician. Positive predictive values (PPV) were calculated for ICD-9 codes. Case-finding algorithms were developed by a classification and regression tree analysis using additional automated data, and these algorithms were compared to true case status. SETTING: Group Health Cooperative. RESULTS: Compared to true cases as determined from chart review, the PPV of ICD-9 codes for NU and MU were 89% (95% CI, 85-92%) and 47% (95% CI, 41-53%), respectively. A higher proportion of true cases (NU: 95%; 95% CI, 90-98%; MU: 56%; 95% CI, 47-66%) were found among subjects with 1+ additional codes occurring in the 12months following the initial code. There was no case-finding algorithm for NU developed given the high PPV of ICD-9 codes. For MU, the best case-finding algorithm classified people as an MU case if they had a fracture in the forearm, hand, or skull and had no visit with an NU diagnosis code in the 12-month post MU diagnosis. PPV for this MU case-finding algorithm increased to 84%. CONCLUSIONS: Identifying NUs with its ICD-9 code is reasonable. Identifying MUs with automated data can be improved by using a case-finding algorithm that uses additional information. Further validation of the MU algorithms in different populations is needed, as well as exploration of its performance in a larger sample.
OBJECTIVE: To evaluate the accuracy of using ICD-9 codes to identify nonunions (NU) and malunions (MU) among adults with a prior fracture code and to explore case-finding algorithms. STUDY DESIGN: Medical chart review of potential NU (N=300) and MU (N=288) cases. True NU cases had evidence of NU and no evidence of MU in the chart (and vice versa for MUs) or were confirmed by the study clinician. Positive predictive values (PPV) were calculated for ICD-9 codes. Case-finding algorithms were developed by a classification and regression tree analysis using additional automated data, and these algorithms were compared to true case status. SETTING: Group Health Cooperative. RESULTS: Compared to true cases as determined from chart review, the PPV of ICD-9 codes for NU and MU were 89% (95% CI, 85-92%) and 47% (95% CI, 41-53%), respectively. A higher proportion of true cases (NU: 95%; 95% CI, 90-98%; MU: 56%; 95% CI, 47-66%) were found among subjects with 1+ additional codes occurring in the 12months following the initial code. There was no case-finding algorithm for NU developed given the high PPV of ICD-9 codes. For MU, the best case-finding algorithm classified people as an MU case if they had a fracture in the forearm, hand, or skull and had no visit with an NU diagnosis code in the 12-month post MU diagnosis. PPV for this MU case-finding algorithm increased to 84%. CONCLUSIONS: Identifying NUs with its ICD-9 code is reasonable. Identifying MUs with automated data can be improved by using a case-finding algorithm that uses additional information. Further validation of the MU algorithms in different populations is needed, as well as exploration of its performance in a larger sample.
Authors: Michael D George; Joshua F Baker; Charles E Leonard; Samir Mehta; Todd A Miano; Sean Hennessy Journal: J Bone Joint Surg Am Date: 2020-07-15 Impact factor: 6.558