Potential Simpson's Paradox in Multicenter Study of IP Chemotherapy for Ovarian Cancer (correspondence/manuscript-open access)

open access (3)

Potential Simpson's Paradox in Multicenter Study of Intraperitoneal Chemotherapy for Ovarian Cancer

 Corresponding author: Alexi A. Wright, MD, Dana-Farber Cancer Institute

To the Editor:

Wright et al¹ used a propensity score matched sample approach to evaluate overall survival and treatment-related toxicities of intraperitoneal (IP)/intravenous (IV) relative to IV-only adjuvant treatment protocols for ovarian cancer.

The authors found substantial differences between institutions in raw IP usage rates. In addition, non-negligible differences between treatment groups were found even after propensity score matching (compare Appendix Table A1 in Wright et al¹) in several variables, including institution at which patients received treatment, Charlson comorbidity scale, age, primary tumor site, and histology. IP treatment was administered to a low percentage of patients in some institutions (4% at institution 3), and to a high percentage in others (67% and 63% at institutions 1 and 2, respectively). Is there a potential for Simpson’s paradox in the authors’ analysis?

Even with propensity matching, patients who received IV-only treatment tended to be older and poorer and have more comorbidities than other patients. Did the matcher try compensating for differences between institutions in percentage of IP treatment by assigning sicker, higher-risk patients receiving IV-only treatment to match patients receiving IP treatment from low IP rate institutions?

Unfortunately, even percentage data on the institutional distribution of propensity score matched patients receiving IV-only versus IP/IV treatment was not presented. More broadly, can we be confident that the observed mortality differences were not largely attributable to differences between institutions—unrelated to the selection of IP/IV versus IV-only treatment? For example, if institutions with a large number of patients receiving IV-only treatment had an overall higher mortality rate, then might this pose a threat of distortion via Simpson’s paradox?

I am concerned that there might have been differences between institutions that were not directly related to the choice of treatment protocol; for example, differences in surgical treatments, differences in patient selection, differences in outpatient and community factors, differences in monitoring approaches, or other hospital factors.

It would be useful to see if there were any substantial differences between centers in overall survival rates (adjusted and/or unadjusted) within a treatment protocol. For example, were the overall survival rates in the IV-only treatment condition comparable across institutions? What was the range of overall survival rates in the IV-only treatment condition across institutions?

Wright et al¹ provide a useful examination of the clinical use of IP/IV chemotherapy at six National Comprehensive Cancer Network institutions. I am hopeful that additional data and analysis can be provided by the authors to further clarify their findings. 

• CORRESPONDENCE: Reply to G.B. Holt

  • Alexi A. Wright,
  • Angel Cronin,
  • and David M. O’Malley

  JCO JCO653782; published online on January 11, 2016;
  • [Full Text]

 In his letter, Holt¹ expressed concern that the survival difference we observed in our study² between intraperitoneal (IP)/intravenous (IV) chemotherapy and IV chemotherapy may be a result of residual confounding between the two groups (eg, differences between institutional practices) rather than because of the treatment itself....................As shown in Table 1, the overall survival was similar between institutions, and the central estimates were in favor of IP/IV for each institution, which is consistent with the aggregate results. In addition, we have included in Table 2 the raw percentages by institution that were omitted from Appendix Table A1 in our original article.² Together, these results suggest that the survival benefit we observed was unlikely to be driven by differences in practices at any individual institution, particularly because the HR that we observed was similar to results from several prior randomized controlled trials.^3-5

REFERENCES

1. 1. Wright AA,
   2. Cronin A,
   3. Milne DE,
   4. et al.

   (2015) Use and effectiveness of intraperitoneal chemotherapy for treatment of ovarian cancer. J Clin Oncol 33:2841–2847.

   FREE Full Text