Response to Zhou and Shan: Interpreting long-term response to rituximab in immune thrombocytopenia

Abstract

We thank Zhou and Shan for their careful reading of our study and for highlighting important questions about the interpretation of registry data on rituximab in immune thrombocytopenia (ITP).¹ We welcome the opportunity to clarify these issues.

First, Zhou and Shan note that inclusion of patients treated early with rituximab depended on subsequent documentation of ITP, potentially underrepresenting patients with durable early responses. As stated in our Methods section, patients were required to have at least two inpatient or outpatient diagnoses of ITP separated by more than 6 months. The criterion was met by a registered ITP diagnosis at an inpatient or outpatient specialist visit. Confirmation of thrombocytopenia was part of diagnostic validation by medical record review, which required at least one documented platelet count below 150 × 10⁹/L.² In the Nordic countries, patients treated with rituximab are routinely followed in specialist care for at least 12 months after treatment, and these visits are registered with an ITP diagnosis irrespective of whether the platelet count has recovered. This practice mitigates the concern that inclusion preferentially captures patients with persistent disease.

Nevertheless, we acknowledge that a patient whose entire disease course was completed within 6 months of the first contact could not have been captured. Such exclusion would, if present, lead to underestimation rather than overestimation of response duration. We cannot determine whether or to what extent this affected our estimated median response duration of 5.4 years.

Second, Zhou and Shan rightly point out that concomitant corticosteroid use may have contributed to the longer response duration observed in the previously untreated group. A meaningful comparison by regimen is, however, difficult: only 32 of the 123 previously untreated patients received rituximab without concomitant corticosteroids, and information on corticosteroid dose was missing or incomplete for a substantial proportion of patients.

In a backward-elimination model, co-administration of corticosteroids was associated with sustained response at 5 years (risk ratio, 8.29; 95% confidence interval, 2.35–29.16) in the non-pretreated group. The wide confidence interval indicates substantial uncertainty and should not be interpreted as clear evidence of a causal effect. Further stratification would yield small subgroups and imprecise estimates, limiting our ability to distinguish the contributions of treatment timing and co-administration of corticosteroids, which the 8-week washout before response assessment does not fully disentangle either.

Third, we agree that the end of the study period in 2018 limited the opportunity for some patients to reach the predefined 2- and 5-year sustained response endpoints (SR2 and SR5). Among the 503 patients who achieved an initial response, all reached the 6-month assessment, so none was excluded from the analysis of initial response. However, 13 patients (2.6%) were still in remission at the end of the study with less than 2 years of follow-up, and 79 (15.7%) had at least two but less than 5 years of follow-up. These patients were included in the analyses but were not counted as sustained responders at time points for which they had insufficient follow-up. Thus, all 503 initial responders remained in the denominator, yielding sustained response proportions of 268 of 503 (53.3%) at 2 years and 158 of 503 (31.4%) at 5 years. This approach yields conservative estimates: restricting the denominators to patients with sufficient follow-up gives 268 of 490 (54.7%) at 2 years and 158 of 411 (38.4%) at 5 years. The risk ratios in Tables 4 and 5 are estimated with the same 503 patients at risk, and we agree that the denominators should have been stated.

Fourth, we agree that complete response is not a pretreatment marker for selecting patients, but rather an early indicator of prolonged response. As stated in the Methods section, the initial response was evaluated separately from the ten baseline characteristics, and our conclusion was that response is largely unpredictable from on routine clinical characteristics. We acknowledge that describing complete response as a predictor in the abstract may have invited the opposite reading. The value of complete response lies where Zhou and Shan place it: in identifying patients likely to stay in remission and in guiding follow-up, not in choosing who should receive rituximab.

Finally, we agree with Zhou and Shan that estimating the causal effect of earlier treatment and predicting response durability are distinct objectives, and we welcome their proposed framework for future studies. However, emulating a target trial requires eligibility, start of treatment, and start of follow-up to fall at the same point in time for every patient, and in our cohort they cannot. Cohort entry requires two ITP contacts more than 6 months apart, a condition that could only be established after treatment had already been given in many patients, so a common time zero would instead have to be defined in all newly diagnosed patients. Upfront rituximab was mainly a Danish practice, so the two strategies were not available in all three countries. The dynamic strategy would require serial platelet counts, which were not systematically available to us. It is a valuable direction for future work, but it would require a different data source from ours.

We thank Zhou and Shan again for their constructive comments, which have helped clarify the interpretation of our findings.

Galina Tsykunova: Writing—original draft; methodology; validation; conceptualization. Eskild Bendix Kristiansen: Methodology; writing—review and editing; validation. Tor Henrik Anderson Tvedt: Writing—review and editing. Henrik Frederiksen: Writing—review and editing; validation. Christian Fynbo Christiansen: Validation; writing—review and editing. Marie Linder: Writing—review and editing. Shahram Bahmanyar: Writing—review and editing. Waleed Ghanima: Writing—review and editing; validation.

Galina Tsykunova: lecture honoraria and advisory board for Janssen, Sanofi, SOBI, and Grifols. Eskild Bendix Kristiansen and Christian Fynbo Christiansen: Department of Clinical Epidemiology receives funding for studies from various companies as research grants to and administered by Aarhus University. Tor Henrik Anderson Tvedt: lecture honoraria and fees for participation in ad boards from Sanofi, Takeda, and Janssen. Henrik Frederiksen: research funding from Sanofi, Novartis, and Alexion. Marie Linder: participation in research projects funded by pharmaceutical companies included Amgen, all regulator-mandated Phase IV studies, all with funds paid to the institution where she is employed (no personal fees). Shahram Bahmanyar: no relevant conflicts of interest. Waleed Ghanima: fees for participation in ad boards from Amgen, Novartis, Pfizer, Principia Biopharma Inc.—a Sanofi Company, Sanofi, SOBI, Grifols, UCB, Argenx, Cellphire, Alpine, Kedrion, HiBio, Hutchmed, and Takeda; lecture honoraria from Amgen, Novartis, Pfizer, Bristol Myers Squibb, SOBI, Grifols, Sanofi, and Bayer; research grants from Bayer, BMS/Pfizer, UCB, Sanofi, SOBI, and Sanofi.

This research received no funding.

The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.