Utility of claims data for identification of date of diagnosis of brain metastases

Brain metastases (BM) affect approximately 10–40% of patients with cancer.¹ Unlike other sites of secondary oncologic involvement, local therapy (ie, radiation and surgery) is commonly utilized for definitive management of BM.² While clinical trials have led to advances in BM management,^3,4 a paucity of population-based studies related to BM exist. Population-based studies are vital for outcomes difficult to assess with prospective or institutionally based retrospective studies, including generalizable assessments of prognosis, patterns of care, and health care disparities. The Surveillance, Epidemiology, and End Results (SEER) database, one of the largest oncologic data sources in the United States, provides no information on BM after diagnosis of the primary malignancy, an important limitation given that many patients develop BM later in their clinical course.⁵ SEER-Medicare data, which link SEER data with Medicare claims and therefore provide clinical information after diagnosis of the primary cancer, have potential for BM investigations but require an accurate date of diagnosis of intracranial involvement. Here, we sought to determine whether claims data could accurately identify the date of intracranial involvement among patients with BM.

We determined the date of BM diagnosis for patients managed at Brigham and Women’s Hospital (BWH) between January 1, 2000 and December 31, 2018 (N = 1383) through manual chart review. We pulled all International Classification of Diseases, Ninth and Tenth Revisions–Clinical Modification (ICD-9/10-CM) codes for secondary malignant neoplasm of the brain, spinal cord, and cerebral meninges (ICD-9-CM 198.3; ICD-10-CM C79.31, 79.32) associated with a patient’s care using the BWH Research Patient Data Registry. We limited our cohort to patients with ≥3 of the above diagnosis codes (N = 1363), an approach associated with 97% sensitivity and 99% specificity for BM identification using claims data⁶ (our sensitivity was 98.6%). We hypothesized that the date of the first BM-associated claim would be within ≤30 days of the BM diagnosis date as identified by manual chart review. Factors associated with poor agreement (>30 days) between the diagnosis date ascertained by claims versus chart review were assessed with a logistic regression model in SAS v9.4. Our study was approved by our institutional review board.

The date of the first BM-associated claim occurred within 15 and 30 days of the true diagnosis date for 1187 (sensitivity: 87.1%) and 1256 patients (sensitivity: 92.1%), respectively. Table 1 depicts characteristics for patients with a claims-based BM diagnosis date within ≤30 days of the actual date versus >30 days. Among covariates in the multivariable model, age, sex, performance status, primary tumor site, presenting symptoms, and number of BM were not predictive of discrepancy between claims versus chart-based diagnosis date (all P > 0.05), but initial treatment with systemic therapy compared with neurosurgery was predictive of this endpoint (odds ratio, 9.1 [95% CI: 4.2–19.5], P < 0.001). The percentage of patients with a claims-based diagnosis date within ≤30 days of the actual diagnosis date was 96.7%, 92.3%, or 78.0% if first managed with neurosurgical resection, brain-directed radiotherapy, or systemic therapy, respectively.

Table 1.

Baseline characteristics of patients with versus without agreement in brain metastasis diagnosis date between claims and manual chart review

	Agreement within 30 Days (n = 1256)	No Agreement within 30 days (n = 107)	P
Age, y, median (interquartile range)	61 (53–68)	58 (51–68)	0.22
Sex, no. (%)			0.36
Female	734 (58)	68 (64)
Male	522 (42)	39 (36)
KPS, no. (%)			0.47
60–80	773 (62)	62 (58)
90–100	483 (38)	45 (42)
Primary cancer type, no. (%)			0.40
Lung	500 (40)	43 (40)
Breast	224 (18)	25 (23)
Melanoma	201 (16)	17 (16)
Other	331 (26)	22 (21)
Initial treatment strategy, no. (%) a			<0.001
Radiation therapy	842 (67)	70 (65)
Surgery	322 (26)	11 (10)
Systemic therapy	92 (7)	26 (24)
No. of brain metastases, median (interquartile range)	2 (1–5)	2 (1–4)	0.55
Neurologic symptoms at diagnosis of brain metastasis, no. (%)	782 (62)	64 (60)	0.61

^aSome percentages do not add up to 100% due to rounding.

Note: Data for presented variables are complete. All covariates above were included in the multivariable logistic model predicting discrepancy between brain metastasis diagnosis date identified via claims versus manual chart review; no additional covariates were included.

We found that billing claims offer a sensitive method to determine BM diagnosis date. We also validated the high sensitivity (>98%) of using ≥3 diagnostic codes in identification of patients with BM.⁶ We included both ICD-9-CM and ICD-10-CM codes, suggesting that our approach remains valid in the current landscape of billing codes. We speculate that routine administration of local therapy in patients with BM facilitates claims-based identification of brain involvement more readily than secondary involvement of other sites.

The primary limitation of our work is that data stemmed from a single institution and may not generalize across institutions. Nevertheless, our study demonstrates the viability of utilizing claims data to identify BM and accurately delineate the diagnosis date in order to facilitate population-based studies related to BM. Caution should be exerted, however, in patients managed with systemic therapy alone.

Conflict of interest statement. Daniel N. Cagney is a recipient of research support from NH Theraguix. Dr Aizer reports research funding from Varian Medical Systems and consulting fees from Novartis. The remaining authors declare no conflicts of interest.

Authorship statement. Study conception: NL, RHB, AAA. Data collection: NL, AAA. Statistical analysis: NL, AAA. Drafting of manuscript: NL, AAA. Critical revision of manuscript: all authors. Supervision: AAA.

Funding

No funding was required for this study.