Journal of NeuroEngineering and Rehabilitation

Beneficiaries who died within three months of the etiological diagnosis were excluded from the cohorts. To be included in the study group, patients were re- quired to have received specified orthotic or pros- thetic services between January 1, 2012 and June 30, 2013. Beneficiaries in the prosthetic sample were re- quired to have a relevant amputation documented in the claims during the study period. This sampling methodology ensured that the database included one year of claims prior to, and at least 18 months fol- lowing, the receipt of the O&P service. Medicare health care claims across all care settings from 2011 to 2014 were obtained for the beneficiaries who met sampling specifications. Care settings included in- patient and outpatient hospitals, long-term care hos- pitals, skilled nursing facilities, inpatient rehabilitation facilities, home health agencies, hospice, physician/ carrier visits, and durable medical equipment, pros- thetics, orthotics, and supplies. This database of study and comparison group bene- ficiaries served as the framework for the analytic sam- ple selected using propensity score matching techniques. We used a one-to-one propensity score match across study and comparison group patients based on etiological diagnosis, comorbidities, patient sociodemographic characteristics (age, gender, race), and historical health care utilization. Additionally, be- cause in the prosthetic analysis the clinical severity (and risk of imminent death) may have been a driver of whether or not the patient received a prosthesis, patients were also matched on the timing of death in relation to amputation, if applicable. As a result, mortality across the groups was excluded as a study outcome for the prosthetic analysis. Propensity score matching techniques are widely used in observational studies when randomized con- trolled trials (RCTs) are not possible or are unethical or impractical to administer [ 15 ] . Literature suggests that applying these techniques to observational stud- ies is an appropriate technique to remove observable selection bias among treatment and comparison groups and can result in findings that look like RCTs [ 16 – 19 ] . In addition, analyses based on administra- tive claims data are much less expensive than clinical trials. Proper matching of the study and comparison group patients limited the number of episodes in- cluded in our study but helped to ensure that the study and comparison group patients were clinically and demographically similar [ 20 ] . Table 1 shows the number of study and comparison group patients in- cluded in each service group before and after match- ing. Propensity score matching resulted in 43,487 matched pairs of Medicare beneficiaries in the lower extremity orthotic model; 34,575 matched pairs in the spinal orthotic model; and 545 matched pairs of re- cent amputees in the prosthetic model. The number of orthotic patients in this current study is higher than in the 2007 – 2010 analysis, a designed increase in sample size resulting from the specifications of the custom cohort database. The relatively small number of beneficiaries included in the lower extremity pros- thetic model was due to the requirement that ampu- tation occur during the study window, which ensured the exclusion of long-term users who received re- placement prosthetics during the study window, and also to the number of variables used in developing the propensity score match. Developing episodes of care Patient episodes were constructed to capture health care diagnoses, utilization, and expenditures prior to and after receipt of the orthotic or prosthetic device. Because actual costs were utilized in the analysis, and because at least one year of claims data prior to and after device provision was included, no additional dis- counting assumptions were incorporated. All patient episodes contained a pre-service window prior to the episode start, which allowed for the identification of comorbid conditions, patterns of institutional care, and other health care utilization used for risk-adjustment during the matching process. Episodes Table 1 Distribution of Pairs (Study Group and Comparison Group Matches) Lower extremity orthotic analysis Spinal orthotic analysis Lower extremity prosthetic analysis Study group Comparison group Study group Comparison group Study group Comparison group Number of patients with O&P service and etiological diagnosis included in custom cohort 239,655 255,156 224,994 240,609 13,823 5959 Number of pairs after propensity score match 43,487 43,487 34,573 34,573 545 545 Percent of patients represented in the effective sample 18.1% 17.0% 15.4% 14.4% 3.9% 9.1% Source: Dobson | DaVanzo analysis of custom cohort Standard Analytic Files (2011 – 2014) for Medicare beneficiaries who received O&P services from January 1, 2012 through June 30, 2013 (and matched comparisons), according to custom cohort database definition Dobson et al. Journal of NeuroEngineering and Rehabilitation 2018, 15 (Suppl 1):55 Page 63 of 72