Identification of cancer chemotherapy regimens and patient cohorts in administrative claims: challenges, opportunities, and a proposed algorithm

Figures & data

Table 1. Percent of patients with diagnoses of interest who also received chemotherapy regimens of interest based on the initial algorithm and expected based on literature.

	Chemotherapy Receipt (%)
Cancer Type	Initial Algorithm	Expected
Breast	15%	>17%^23
Cervical	<1%	>32%^24
Colorectal	1%	>9%^18
Lung	20%	>18%^23
NHL	16%	>82%^23
Ovarian	33%	>63%^25
Pancreatic	<1%	>43%^26
Testicular	13%	>6%^27
Uterine	<1%	>7%^28

Abbreviations. NHL, non-Hodgkin lymphoma.

Table 2. Description and rationale of iterative revisions to identify patient cohorts.

	Description of Change	Rationale
Revision 1	Reverse order of first two inclusion criteria	Eliminated patients screened but who do not have cancer Retained patients with initial cancer diagnosis outside 12-month look-back period
	Chemotherapy claim on ≥1 day assumed to receive all doses	Timing of claims did not necessarily reflect clinical care (e.g. colorectal patients with claim for 5-FU on ≥1 day assumed to have received a 2-day regimen in FOLFOX)
	Oral prednisone and methotrexate not used for case identification or exclusion	Multi-indication treatments that did not influence FN risk were not included Patients may fill oral prescription days to weeks prior to infusion
	Remove cisplatin + vinorelbine in lung cancer	Removed - not commonly used in clinical practice per clinician advisement
	Remove BEP regimen in testicular cancer	Removed - potential interaction due to bleomycin and G-CSF per clinician advisement
Revision 2	G-CSF treatment window	Patients may fill outpatient G-CSF prescriptions separately, so allowing 7 days before and after infusion allows for a more accurate capture of G-CSF use
	Prior cancers exclusion	This exclusion was too narrow and eliminated patients even with early-stage or other cancers without G-CSF or chemotherapy exposure
	Categorize on most recent diagnosis	To improve the likelihood that patients are placed in the disease cohort that most likely reflects their chemotherapy and G-CSF use
Revision 3	5-day window to identify chemotherapy drugs	More accurately capture chemotherapy use, accounting for differences in the timing of claims vs. clinical care
	Prior chemotherapy exclusion	Exclusions were done after initial data extraction
	Separate exclusion for prior radiation therapy	Moved to earlier in the exclusion process to streamline data extraction
	Include paclitaxel regimens for lung cancer	Were not always identified; the issue was resolved when the 5-day window for regimen identification was implemented
Revision 4	Metastatic disease or bone-targeting agent exclusion ≤183 days	Exclusions were done after initial data extraction

Table 3. Waterfall describing the number of patients included after applying inclusion and exclusion criteria to data from one Research Partner.

A. Initial Algorithm		B. Revised Algorithm
Inclusion Criteria	No. Included		Revision 1	Revision 2	Revision 3	Revision 4
1 – cancer diagnosis	22,859	1 – chemotherapy	5,320	5,182	13,600^a	13,600^a
2 – chemotherapy	2,814	2 – cancer diagnosis	3,645	3,396	6,010	6,010
3 – G-CSF claim	1,318	3 – G-CSF claim	1,501	1,476	1,779	1,801
4 – 365 days enrollment	843	4 – 365 days enrollment	931	916	1,093	1,116
5 – age 20 years or older	842	5 – age 20 years or older	930	915	1,086	1,109
Exclusion Criteria
--	--	1a – prior radiation	Not Excl.	Not Excl.	1,081	1,068
1 – prior cancer	781	1b – prior cancer	853	Not Excl.	Not Excl.	Not Excl.
2 – prior chemotherapy	631	2 – prior chemotherapy	702	727	Not Excl.	Not Excl.
3 – prior G-CSF	610	3 – prior G-CSF	668	689	962	958
4 – prior SNF or hospice	610	4 – prior SNF or hospice	668	689	962	958
5 – clinical exclusions	420	5 – clinical exclusions	449	462	620	886
Final Cohort	420					886

(A) First data run; (B) Results of each incremental revision to the cohort identification algorithm.

Abbreviations. G-CSF, granulocyte-colony stimulating factor; SNF, skilled nursing facility.

^aAll records within a 5-day window; individuals may contribute more than one record.

Figure 1. Final algorithm to identify cohorts of patients receiving multi-agent chemotherapy.

Cancer Treatment & Survivorship Facts & Figures 2019-2021. 2022. https://www.cancer.org/content/dam/cancer-org/research/cancer-facts-and-statistics/cancer-treatment-and-survivorship-facts-and-figures/cancer-treatment-and-survivorship-facts-and-figures-2019-2021.pdf

 Google Scholar

Levinson KL, Bristow RE, Donohue PK, et al. Impact of payer status on treatment of cervical cancer at a tertiary referral center. Gynecol Oncol. 2011;122(2):324–327.

 PubMed Web of Science ®Google Scholar

Miller KD, Nogueira L, Mariotto AB, et al. Cancer treatment and survivorship statistics, 2019. CA A Cancer J Clin. 2019;69(5):363–385.

 PubMed Web of Science ®Google Scholar

Cancer Trends Proress Report. 2021. https://progressreport

 Google Scholar

Doleh Y, Lal LS, Blauer-Petersen C, et al. Treatment patterns and outcomes in pancreatic cancer: retrospective claims analysis. Cancer Med. 2020;9(10):3463–3476.

 PubMed Web of Science ®Google Scholar

Leveridge MJ, Siemens DR, Brennan K, et al. Temporal trends in management and outcomes of testicular cancer: a population-based study. Cancer. 2018;124(13):2724–2732.

 PubMed Web of Science ®Google Scholar

Zakem SJ, Robin TP, Smith DE, et al. Evolving trends in the management of high-intermediate risk endometrial cancer in the United States. Gynecol Oncol. 2019;152(3):522–527.

 PubMed Web of Science ®Google Scholar

Data availability statement

None.