Table 2.
Regression discontinuity design analysis of the effect of increased utilization of chemotherapy on all cause mortality in women with advance ovarian cancer
| Model* | Year range | New England and east south central census divisions† | South Atlantic, west north central, and east north central census divisions‡ | P value§ | |||
|---|---|---|---|---|---|---|---|
| Hazard ratio (95% CI) | Participants¶ | Hazard ratio (95% CI) | Participants¶ | ||||
| 1 | 2011-12 | 0.82 (0.76 to 0.89) | 1156 | 1.00 (0.95 to 1.05) | 4836 | <0.001 | |
| 2 | 2007-12 | 0.81 (0.71 to 0.94) | 3014 | 1.02 (0.93 to 1.12) | 15 400 | 0.001 | |
Relative hazards of death from any cause among women treated in 2012 compared with prior years were estimated with Cox proportional hazard models. Model 1 estimates the relative hazards of diagnosis in 2012 compared with 2011, ignoring mortality time trends. Model 2 estimates the relative hazard of diagnosis in 2012 compared with prior years, adjusting for trends in mortality.
New England and east south central census division experienced a discontinuous increase in the frequency of women treated with neoadjuvant chemotherapy between 2011 and 2012. Women treated in these regions in 2012 had 41% greater odds of receiving neoadjuvant chemotherapy compared with prior years.
South Atlantic, west north central, and east north central census division are considered negative controls because the frequency of neoadjuvant chemotherapy in these regions did not change between 2011 and 2012.
P values were obtained from Wald tests comparing relative hazards between rapidly adopting regions and controls in a Cox proportional hazard difference-in-differences models.
Survival information is missing for one patient from New England and east south central census divisions, and two patients from South Atlantic, west north central, and east north central census divisions treated in 2011 and 2012.