Association of COVID-19 Vaccination With SARS-CoV-2 Infection in Patients With Cancer

This cohort study assesses the association between SARS-CoV-2 vaccination and SARS-CoV-2 infections among a population of Veterans Affairs (VA) patients with cancer.

Treatment timing prior to vaccination (if in the vaccinated cohort) or prior to entry date (if in the unvaccinated cohort) was defined based on pharmacy records in the CDW. Four treatment timing categories were defined: (1) Distant treatment; (2) Recent treatment; (3) Current treatment; (4) Treatment after vaccine. Regarding category (1), note that all patients in the study population received cancer-directed therapy on or after August 15, 2010, but this treatment ceased at least six months prior to the date of vaccination or matching entry date. The patients in category (2) had received their last dose of systemic therapy sometime in the period between three and six months prior to vaccination or matching. Patients in category (2) therefore represent a group that received systemic therapy more recently than the patients in category (1) but are no longer on active therapy at the time of vaccination or matching. Regarding category (3), patients were included if a patient received any systemic therapy within three months prior to vaccination or matching. Patients in category (4) received their first dose of systemic therapy after vaccination or matching but prior to study end date of May 4, 2021.
Treatment type was defined for the patients on current therapy at the time of vaccination (if in the vaccinated cohort) or entry date (if in the unvaccinated cohort). Patients were considered to be on current therapy if a patient received any systemic therapy within three months prior to vaccination or matching. All agents administered within the three month window were considered. Individual drugs were classified as chemotherapy, immunotherapy, targeted therapy, or endocrine therapy using the mapping in eTable 13. If all the agents given to a patient in a three month period were the same type, that patient was assigned to that treatment type category. Regimens consisting of combinations of treatment types were prioritized in the following order: If any chemotherapy agent was administered in combination with another agent (e.g., chemoimmunotherapy), it was considered a chemotherapy-containing regimen. If immunotherapy agent was administered in combination with another agent that was not chemotherapy, it was considered an immunotherapy regimen. Targeted regimens contain targeted agents, but not chemotherapy or immunotherapy. Endocrine regimens were composed exclusively of endocrine therapy agents. All routes of cancer treatment were included, including oral, intravenous, intramuscular and subcutaneous.
Cancer type was determined from malignancy ICD-10 codes by the following algorithm for each patient (eFigure 6). For each patient, the most recent malignancy ICD-10 code(s) associated with a cancer-directed systemic treatment was identified. If there was a unique ICD-10 associated with that encounter, then the cancer type associated with that code was used. However, if there were multiple ICD-10 codes, and the ICD-10 codes were associated with different cancer types, ties were broken in with the following stepwise algorithm. If one of the ICD-10 codes were for malignancy of brain, bone, or lymph node, the primary site was assumed to be the remaining code(s) since brain, bone, and lymph nodes are common metastatic sites. For example, if a patient was identified by both a colorectal malignancy-associated ICD-10 code and a bone malignancy-associated ICD-10 code, the colorectal malignancy was assumed to be the primary site. If there were still a tie, the primary malignancy was decided by majority vote of the malignancy ICD-10 codes over the past 6 months prior to the most recent treatment-associated malignancy ICD-10 code. Remaining ties were broken via cancer type and treatment matching using a cancer type-treatment matrix. For example, if a patient received antiandrogen therapy and had a prostate malignancy-associated ICD-10 code, that patient was assumed to have prostate cancer. If the patient received a therapy associated with more than one cancer type, the patient was assumed to have the more common cancer type based on SEER national cancer prevalence statistics 3 . If a patient did not receive any of the typical treatments associated with their malignancy ICD-10 code, that patient was assigned to the other malignancy ICD-10 code. Remaining ties were classified as unknown.
Cancer category was determined using the cancer type as defined above. Hematologic malignancies were defined as essential thrombocythemia, leukemia/myelodysplastic syndrome (MDS)/myelofibrosis (MF), multiple myeloma, mastocytoma, and polycythemia vera. Solid malignancies were defined as brain, breast, colorectal, connective and soft tissue, squamous cell carcinoma of the head and neck (SCCHN)/cutaneous SCC (CSCC)/skin, esophagus/gastric, gynecologic malignancies (Gyn), hepatocellular carcinoma (HCC), lung, lymphoma, melanoma, neuroendocrine, anal/biliary/GIST/pancreas/small intestine (other GI), prostate, renal cell carcinoma (RCC), and urothelial cancer. ICD codes classified as "other/unknown" was split among the solid and hematologic malignancies based on the specific ICD code.
Total death was defined as any death recorded in the CDW occurring on the day of or in the four weeks following SARS-CoV-2 infection or a censoring event. Example: Patient 1, a vaccinated patient was vaccinated with their first dose on December 15 th , 2020, infected with COVID-19 on January 1 st 2021, and died on January 21 st 2021. Patient 2, patient 1's unvaccinated matched control, was vaccinated on December 20 th 2020. Given that patient 2 was vaccinated on December 20 th 2020, both Patient 1 and Patient 2 would be censored on that day (Patient 2 censored as control and re-enters study as vaccinated). Although Patient 1 did indeed experience a death, this patient would not be counted in the total death because January 21 st 2021 is more than 28 days past Patient 1's censored date. COVID-19 related death was defined as any death recorded in the CDW within four weeks after documentation of SARS-CoV-2 infection, excluding patients who were censored from the study population prior to infection 4 . In the example above, although Patient 1's death was within 28 days of a SARS-COV-2 infection, the death would not be counted as a COVID-19 related death because Patient 1 was censored. On the other hand, if Patient 2 (Patient 1's matched control) was vaccinated on January 5 th 2021, then Patient 1's death would be counted as a COVID-19 related death.

Matching Algorithm
Vaccinated patients were matched with unvaccinated patients on predictors of SARS-COV-2 infection and vaccination. These variables included age, race/ethnicity, VA facility, rurality of home address, cancer type, and treatment type/timing. Age was matched with minimum distance matching and all other variables were matched with exact matching. So, a non-Hispanic white veteran with a rural home address and prostate cancer on endocrine therapy at the Boston VA will only be matched with another non-Hispanic white veteran with a rural home address and prostate cancer on endocrine therapy at the Boston VA. If more than one control patient matches the vaccinated patient, the control closest in age to the vaccinated patient is chosen. If there are no controls available that exactly match the vaccinated patient, the unmatched vaccinated patient is discarded. Matching on additional variables (gender, comorbidity index, BMI) was not included, as their inclusion decreased sample size but did not improve matching quality.
During this matching process, patients who received vaccination after the matching date were eligible to be matched as an unvaccinated control. Excluding these patients would create a bias, since people who are symptomatic from early infection may avoid getting vaccinated and thereby bias the unvaccinated cohort towards a higher early COVID infection rate. If an unvaccinated control was subsequently vaccinated, follow-up for both the control and corresponding case was censored on the date of the control's vaccination. After the control's vaccination, the control would cross over to the vaccinated cohort and would be matched to a new unvaccinated control.
To assess for residual confounding, and thereby evaluate matching quality, as in the prior study 5 we examined the cumulative incidence of infection in the first 14 days of follow-up for both vaccinated and controls. This provides a "negative control" in regard to confounding. During this initial period, vaccinated and unvaccinated groups are expected to exhibit similar rates of infection if exchangeability between the groups holds, i.e., if there is no confounding.

Statistical Analysis
For first-dose effectiveness analyses, the index date for each vaccinated patient was set to their vaccination date, and the index date for their matched control was set to this same date. For analyses evaluating effectiveness after the second dose, date of second vaccination was used as the index date for each vaccinated patient and their matched control. For second-dose analyses, matched pairs where (1) the vaccinated patient did not have a second dose, (2) either member was infected before the second-dose index date, (3) the control died before the second-dose index date, or (4) the control was vaccinated before the second-dose index date were excluded.
Curves of cumulative incidence of infection for vaccinated patients and unvaccinated controls were generated using a Fine-Gray adjusted proportional subdistribution hazards model accounting for the competing risk of death 6 , and adjusting for all variables used in matching except for VA facility and cancer type, as these variables introduced too many levels. The Fine-Gray subdistribution hazard function was used to measure risk. These measures were evaluated over pre-defined intervals analogous to those used in the original Pfizer BNT162b2 clinical trial 7 : day 0 after the first dose of vaccine to end of study, day 0 after the first dose to date of second dose, day 0 to day 13 after the second dose and day 14 after the second dose to end of follow-up. As protective immunity is expected to be minimal in the first two weeks after vaccination, a secondary analysis was performed considering the interval day 14 after the first dose to the date of the second dose. Confidence intervals for the effectiveness and cumulative incidence curves were calculated using percentile bootstrapping. Analyses were performed using R version 4.0.2.