Quality, Spending and Outcomes in Women With Breast Cancer
We identified 15357 women with stage 0-III breast cancer who were eligible for at least one quality measure (Table 1). As expected, most were non-Hispanic white (81.7%), few had comorbidity scores of 2 or greater (7.4%), a majority had negative nodes (58.5%), and most had HR-positive disease (63.9%). Table 2 lists each quality measure with its number of eligible patients and overall concordance. The median number of eligible patients per measure was 529 (interquartile range [IQR] = 366–1700), and the median concordance was 76.9% (IQR = 64.5%–95.3%). Concordance relative to recommended-therapy measures was lower than concordance relative to nonrecommended therapy measures (median 73.4% vs 95.1%), but this difference was not statistically significant (P = .13). The five-year overall survival was 87.5%.
There were 94 HSAs with 25 or more eligible patient encounters. Only 4.5% of the cohort came from an HSA with less than 25 patients. Those for whom no HSA could be assigned were grouped into a separate region. In total 99 regions were defined, with a median of 85 eligible patients per region (IQR = 47–158). Regions demonstrated substantial variability in overall concordance, five-year survival, and median per patient expenditure (Figure 1). There were no statistically significant trends in overall concordance across spending quintiles, and there were no statistically significant trends in outcomes across either spending or quality quintiles (Figure 2). After disaggregating quality and spending into their component parts, we found that median part B expenditure demonstrated a statistically significant positive correlation with recommended-therapy concordance (Spearman correlation = 0.22, P = .027) and a statistically significant negative correlation with nonrecommended-therapy concordance (Spearman correlation = -0.29, P = .004) (Figure 3). Concordance relative to recommended- and unnecessary-therapy measures did not correlate with each other (P = .36) (Figure 4) or with overall survival (P = .42).
(Enlarge Image)
Figure 1.
Distribution of concordance telative to 27 quality measures (blue), five-year overall survival (red), and total expenditure in the year after diagnosis (green) across 99 Regions. Concordance and overall survival plotted relative to the left axis; expenditures plotted relative to the right axis. For each data element, the minimum, mean, and maximum values are displayed; median values for the three categories were similar to the means (82%, 90%, and $20039, respectively). The "X" symbols denote +/- one standard deviation (for concordance and for survival) or the interquartile range (for expenditure).
(Enlarge Image)
Figure 2.
Concordance relative to 27 quality measures (blue), five-year overall survival (red), and median total expenditure in the year after diagnosis (green) across expenditure (top) and quality (bottom) quintiles. Concordance and overall survival plotted relative to the left axis; expenditures plotted relative to the right axis.
(Enlarge Image)
Figure 3.
Scatter plot of each region's median part B expenditure in the tear after diagnosis vs its concordance relative to 20 for-treatment (upper) and seven against-treatment (lower) measures. Spearman correlation coefficients were 0.22 (P = .027) for the upper and -0.29 (P = .004) for the lower plots (lines represent linear best-fit trends).
(Enlarge Image)
Figure 4.
Scatter plot of each tegion's concordance relative to 20 for-treatment measures vs its concordance relative to seven against-treatment measures. Each diamond represents the intersection of an health care service area's concordance relative to against-treatment (x-axis) and for-treatment (y-axis) measures. Cross hairs highlight the mean values for the respective measure-types across all health care service areas. Regions in the top right demonstrate higher quality for both types of measures; regions in the bottom left demonstrate lower quality for both types of measures; regions in the top left tend to provide treatment regardless of the recommendation (higher concordance relative to for-treatment measures and lower concordance relative to against-treatment measures); regions in the bottom right tend not to provide treatment regardless of the recommendation.
Two multivariate models were created to identify independent predictors of part B expenditures: one focused on concordance relative to recommended-therapy measures and the other on concordance relative to unnecessary-therapy measures. For each 1% increase in performance relative to recommended-therapy measures, part B expenditure increased $280 (95% CI = $104 to $456, P < .001). For each 1% increase in performance relative to nonrecommended-therapy measures part B expenditure decreased $136 (95% CI = $41 to $232, P < .001). Median part A expenditures were not correlated with either recommended or nonrecommended-therapy measure concordance (P = .56 and .44, respectively).
After classifying regions into four groups based on their performance relative to the median recommended- and unnecessary-therapy concordance values, we found that regions with quality less than the median for both demonstrated 22% greater part A expenditure than regions with quality more than the median for both; part B expenditures were similar ($22004 vs $22017) (Figure 5; neither difference was statistically significant). Regions that provided less treatment regardless of the type of measure demonstrated the lowest part B expenditure (P ≤ .001) and the lowest five-year overall survival (NS).
(Enlarge Image)
Figure 5.
Medicare expenditures and five-year overall survival for HSA-based regions defined by their concordance relative to 20 for- and seven against-treatment breast cancer quality measures. The groups are 1) higher quality relative to both for- and against-treatment measures, 2) more use of therapy regardless of the recommendation, 3) less use of therapy regardless of the recommendation, and 4) lower quality relative to both for- and against-treatment measures. The Kruskal-Wallis test showed that only mean part B expenditure demonstrated statistically significant variation across the four groups (two-sided P < .01).
Results
We identified 15357 women with stage 0-III breast cancer who were eligible for at least one quality measure (Table 1). As expected, most were non-Hispanic white (81.7%), few had comorbidity scores of 2 or greater (7.4%), a majority had negative nodes (58.5%), and most had HR-positive disease (63.9%). Table 2 lists each quality measure with its number of eligible patients and overall concordance. The median number of eligible patients per measure was 529 (interquartile range [IQR] = 366–1700), and the median concordance was 76.9% (IQR = 64.5%–95.3%). Concordance relative to recommended-therapy measures was lower than concordance relative to nonrecommended therapy measures (median 73.4% vs 95.1%), but this difference was not statistically significant (P = .13). The five-year overall survival was 87.5%.
There were 94 HSAs with 25 or more eligible patient encounters. Only 4.5% of the cohort came from an HSA with less than 25 patients. Those for whom no HSA could be assigned were grouped into a separate region. In total 99 regions were defined, with a median of 85 eligible patients per region (IQR = 47–158). Regions demonstrated substantial variability in overall concordance, five-year survival, and median per patient expenditure (Figure 1). There were no statistically significant trends in overall concordance across spending quintiles, and there were no statistically significant trends in outcomes across either spending or quality quintiles (Figure 2). After disaggregating quality and spending into their component parts, we found that median part B expenditure demonstrated a statistically significant positive correlation with recommended-therapy concordance (Spearman correlation = 0.22, P = .027) and a statistically significant negative correlation with nonrecommended-therapy concordance (Spearman correlation = -0.29, P = .004) (Figure 3). Concordance relative to recommended- and unnecessary-therapy measures did not correlate with each other (P = .36) (Figure 4) or with overall survival (P = .42).
(Enlarge Image)
Figure 1.
Distribution of concordance telative to 27 quality measures (blue), five-year overall survival (red), and total expenditure in the year after diagnosis (green) across 99 Regions. Concordance and overall survival plotted relative to the left axis; expenditures plotted relative to the right axis. For each data element, the minimum, mean, and maximum values are displayed; median values for the three categories were similar to the means (82%, 90%, and $20039, respectively). The "X" symbols denote +/- one standard deviation (for concordance and for survival) or the interquartile range (for expenditure).
(Enlarge Image)
Figure 2.
Concordance relative to 27 quality measures (blue), five-year overall survival (red), and median total expenditure in the year after diagnosis (green) across expenditure (top) and quality (bottom) quintiles. Concordance and overall survival plotted relative to the left axis; expenditures plotted relative to the right axis.
(Enlarge Image)
Figure 3.
Scatter plot of each region's median part B expenditure in the tear after diagnosis vs its concordance relative to 20 for-treatment (upper) and seven against-treatment (lower) measures. Spearman correlation coefficients were 0.22 (P = .027) for the upper and -0.29 (P = .004) for the lower plots (lines represent linear best-fit trends).
(Enlarge Image)
Figure 4.
Scatter plot of each tegion's concordance relative to 20 for-treatment measures vs its concordance relative to seven against-treatment measures. Each diamond represents the intersection of an health care service area's concordance relative to against-treatment (x-axis) and for-treatment (y-axis) measures. Cross hairs highlight the mean values for the respective measure-types across all health care service areas. Regions in the top right demonstrate higher quality for both types of measures; regions in the bottom left demonstrate lower quality for both types of measures; regions in the top left tend to provide treatment regardless of the recommendation (higher concordance relative to for-treatment measures and lower concordance relative to against-treatment measures); regions in the bottom right tend not to provide treatment regardless of the recommendation.
Two multivariate models were created to identify independent predictors of part B expenditures: one focused on concordance relative to recommended-therapy measures and the other on concordance relative to unnecessary-therapy measures. For each 1% increase in performance relative to recommended-therapy measures, part B expenditure increased $280 (95% CI = $104 to $456, P < .001). For each 1% increase in performance relative to nonrecommended-therapy measures part B expenditure decreased $136 (95% CI = $41 to $232, P < .001). Median part A expenditures were not correlated with either recommended or nonrecommended-therapy measure concordance (P = .56 and .44, respectively).
After classifying regions into four groups based on their performance relative to the median recommended- and unnecessary-therapy concordance values, we found that regions with quality less than the median for both demonstrated 22% greater part A expenditure than regions with quality more than the median for both; part B expenditures were similar ($22004 vs $22017) (Figure 5; neither difference was statistically significant). Regions that provided less treatment regardless of the type of measure demonstrated the lowest part B expenditure (P ≤ .001) and the lowest five-year overall survival (NS).
(Enlarge Image)
Figure 5.
Medicare expenditures and five-year overall survival for HSA-based regions defined by their concordance relative to 20 for- and seven against-treatment breast cancer quality measures. The groups are 1) higher quality relative to both for- and against-treatment measures, 2) more use of therapy regardless of the recommendation, 3) less use of therapy regardless of the recommendation, and 4) lower quality relative to both for- and against-treatment measures. The Kruskal-Wallis test showed that only mean part B expenditure demonstrated statistically significant variation across the four groups (two-sided P < .01).