Development of population-level colon cancer pathway concordance measures and association with survival.

Clinical cancer pathways help standardize healthcare delivery to optimize patient outcomes and health system costs. However, population-level measurement of concordance between standardized pathways and actual care received is lacking. Two measures of pathway concordance were developed for a simplified colon cancer pathway map for Stage II-III colon cancer patients in Ontario, Canada: a cumulative count of concordant events (CCCE) and the Levenshtein algorithm. Associations of concordance with patient survival were estimated using Cox proportional hazards models adjusted for patient characteristics and time-dependent cancer-related activities. Models were compared and the impact of including concordance scores was quantified using the likelihood ratio chi-squared test. The ability of the measures to discriminate between survivors and decedents was compared using the C-index. Normalized concordance scores were significantly associated with patient survival in models for cancer stage-a 10% increase in concordance for Stage II patients resulted in a CCCE score adjusted hazard ratio (aHR) of death of 0.93, 95% CI 0.88-0.98 and a Levenshtein score aHR of 0.64, 95% CI 0.60-0.67. A similar relationship was found for Stage III patients-a 10% increase in concordance resulted in a CCCE aHR of 0.85, 95% CI 0.81-0.88 and a Levenshtein aHR of 0.78, 95% CI, 0.74-0.81. Pathway concordance can be used as a tool for health systems to monitor deviations from established clinical pathways. The Levenshtein score better characterized differences between actual care and clinical pathways in a population, was more strongly associated with survival and demonstrated better patient discrimination.

Forecasting Ontario Oncology Drug Expenditures: A Hybrid Approach to Improving Accuracy.

BACKGROUND: The Provincial Drug Reimbursement Program (PDRP) at Cancer Care Ontario (CCO) is responsible for monitoring actual and projected outpatient intravenous cancer drug spending in the province. We developed a hybrid forecasting approach combining automated time-series forecasting with expert-customizable input. OBJECTIVE: Our objectives were to provide a flexible tool in which to incorporate multiple forecasts and to improve the accuracy of the resulting forecast. METHODS: The approach employed linear and non-linear time-series techniques and a combined hybrid model incorporating both approaches. We developed an interactive tool that incorporated the statistical models and identified the best performing forecast according to standard goodness-of-fit measures. Model selection procedures considered both the amount of historical expenditure data available per drug policy and the individual policy contributions to the overall budget. The user was allowed to customize forecasts based on knowledge of external factors related to policy or price changes and new drugs that come to market RESULTS: A comparison of 2016/17 fiscal year expenditures showed that all policies with a significant contribution to the overall budget were forecast with < 4% error. Forecasting error was reduced by at least $Can5 million for the nine most expensive policies compared with expert opinion. This approach to drug budget forecasting was implemented in Ontario for the first time in the 2017/18 fiscal year, where 1% error was observed for the overall budget, corresponding to an overestimate of expenditures by $Can3.0 million. CONCLUSION: We introduced a pragmatic approach for regular forecasting by budget holders in Ontario. Our approach to isolating 'big budget' from 'small budget' drugs using an 80-20 rule and providing multiple forecasts depending on the length of the drug expenditure histories is transferable to other jurisdictions.