Utility of claims data for delineation of intracranial treatment among patients with brain metastases

Brain metastases (BM) affect approximately 30% of patients with cancer.¹ Given the guarded intracranial efficacy of many systemic therapies, BM are routinely managed with local therapies, including brain-directed radiation and neurosurgery.^2,3 Population-based studies of BM have historically been lacking, largely due to uncertainty regarding whether claims can reliably (i) identify BM, (ii) delineate the date of intracranial involvement, and (iii) characterize intracranial management. Although validated approaches now exist to identify BM⁴ and the associated date of diagnosis,⁵ no validated claims-based methods to characterize intracranial management exist. We sought to assess whether claims can accurately identify the receipt and date of brain-directed local therapies among patients with BM.

We identified Medicare beneficiaries with BM initially managed with whole-brain radiotherapy (WBRT, n = 110), stereotactic radiosurgery/radiotherapy (SRS/SRT, n = 216), and/or neurosurgery (n = 88) at Brigham and Women’s Hospital/Dana-Farber Cancer Institute (BWH/DFCI) between January 2007 and December 31, 2016 through manual chart review and created a clinically annotated database of such patients. Then, using the BWH/DFCI Medicare-claims database, we linked relevant International Classification of Diseases (ICD) revisions 9/10, Current Procedural Terminology v4, and/or Healthcare Common Procedure Coding System codes for brain-directed non-stereotactic radiation (at BWH/DFCI, such treatment comprises WBRT), brain-directed SRS/SRT, and neurosurgery to our clinically annotated database in order to assess the sensitivity and specificity of claims data in delineation of intracranial management. Of note, only radiation or neurosurgery-related codes linked to an ICD9–Clinical Modification (CM) or ICD10-CM diagnosis code for brain metastases (ICD-9: 198.3; ICD-10: C79.31–79.32) qualified as brain-directed treatment. Sensitivity investigations were limited to initial intracranial management; specificity delineations extended to the entire clinical course. Moreover, in an attempt to capture those patients who actually underwent treatment, codes indicating delivery of treatment were utilized, while codes for simulation or planning, for example, were not (Supplementary Table 1). Factors associated with poor sensitivity were assessed via univariable logistic regression. The small number of events prohibited multivariable modeling. Our study was institutional review board–approved. SAS v9.4 was used for all analyses.

The claims-based approach was 94%, 98%, and 94% sensitive for identifying receipt of WBRT, SRS/SRT, and neurosurgery, respectively, relative to manual chart review (Table 1). Of patients with treatment claims present, the date of the first treatment-related claim occurred within 15 days of the actual treatment date for 96%, 93%, and 96% of patients receiving WBRT, SRS/SRT, and neurosurgery, respectively. When limiting to patients who never received WBRT (n = 135), SRS/SRT (n = 110), or neurosurgery (n = 225), these treatment codes demonstrated 98%, 97%, or 99% specificity, respectively (Table 1). On univariable regression of patients receiving local treatment based upon chart review, sex, Charlson comorbidity score, primary tumor site, seizures at BM diagnosis, and number of extracranial metastases were not predictive of a claims-based versus chart-based discrepancy in identification of treatment strategy (P > 0.05). Covariates associated with poorer claims-based treatment identification among patients managed with WBRT and SRS/SRT included older age and lower performance status, respectively; in both cohorts, the absence of symptoms at BM diagnosis and earlier year of treatment were also associated with poorer concordance between claims and chart-based review (P < 0.05).

Table 1.

Sensitivity and specificity of claims-based identification of intracranial treatment strategy for patients with BM compared with manual chart review

	Patients Receiving WBRT per Chart Review (N = 110)a	Patients Receiving SRS/SRT per Chart Review (N = 216)a	Patients Undergoing Neurosurgical Resection per Chart Review (N = 88)a
Presence of treatment-related claims	103	211	83
Absence of treatment- related claims	7	5	5
Sensitivity of claims	94%	98%	94%
	No WBRT per chart review (N = 135)  b	No SRS/SRT per chart review (N = 110)  b	No neurosurgical resection per chart review (N = 225)  b
Presence of treatment-related claims	3	3	3
Absence of treatment- related claims	132	107	222
Specificity of claims	98%	97%	99%

^aAs initial management of brain metastases.

^bDuring a patient’s entire clinical course.

In this validation study, we found that claims were highly sensitive and specific for delineating intracranial management strategy among patients with BM, with sensitivities of 94%, 98%, and 94% and specificities of 98%, 97%, and 99% for WBRT, SRS/SRT, and neurosurgery, respectively. The dates of such claims were also highly accurate, to within 15 days, in 96%, 93%, and 96% of cases, respectively. This high sensitivity indicates that brain-directed treatment can be reliably identified via claims, while the high specificity suggests that treatment-related claims are not used extraneously and are generally only present for patients undergoing treatment.

The management of patients with BM is complex given the heterogeneity of the patient population with respect to age, performance status, primary tumor site, intracranial and systemic disease burden, prior treatment, and patient preference.⁶ The methodology presented here will facilitate patient-level, systems-level, and population-level characterization of treatment patterns among patients with BM, delineation of predictors of such treatment, and assessments of the impact of intracranial management on population-level outcomes. Despite the limitation that data stemmed from a single institution, it seems plausible that intracranial management can now be identified using claims data, thereby expanding the reach of population-based investigations to impact patients with BM.

Funding

No funding was required for this study.

Conflict of interest statement.

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: all authors. Data collection: NL, AAA. Statistical analysis: NL, PC, AAA. Drafting of the manuscript: NL, AAA. Critical revision of the manuscript: all authors. Supervision: AAA.

Data availability statement.

Baseline characteristics for patients and logistic regression output will be shared on reasonable request to the corresponding author pending institutional approval.