Histogram of 1-year and 5-year cumulative costs. Frequency is the number of patients reporting costs. WUHNCCI indicates Washington University Head and Neck Cancer Comorbidity Index.
Marginal effect of Washington University Head and Neck Cancer Comorbidity Index on 1-year (A) and 5-year (B) cumulative costs.
Hollenbeak CS, Stack BC, Daley SM, Piccirillo JF. Using Comorbidity Indexes to Predict Costs for Head and Neck Cancer. Arch Otolaryngol Head Neck Surg. 2007;133(1):24-27. doi:10.1001/archotol.133.1.24
Copyright 2007 American Medical Association. All Rights Reserved. Applicable FARS/DFARS Restrictions Apply to Government Use.2007
To determine whether the general Charlson Comorbidity Index (CCI) and the head and neck cancer–specific Washington University Head and Neck Cancer Comorbidity Index (WUHNCCI) were useful for predicting cost of treatment for elderly patients with head and neck cancer.
Retrospective, observational study.
A total of 1780 Medicare patients with head and neck cancer, who were treated between 1984 and 1994, were analyzed using the Surveillance, Epidemiology, and End Results (SEER)-Medicare–linked database.
Main Outcome Measures
Total Medicare payments were accumulated for each patient up to 1 and 5 years. Linear regression was used to estimate the impact of the comorbidity indexes on costs, controlling for demographics, site, stage, and treatment modality.
Neither the WUHNCCI nor the CCI was significantly associated with 1-year costs. However, the effect of the WUHNCCI on 5-year costs was statistically significant (P<.001). A 1-point increase in the WUHNCCI from 4 to 5 was associated with an increase in 5-year costs of $2105. A 1-point increase in the WUHNCCI from 9 to 10 was associated with an increase in 5-year costs of $2837.
These results suggest that comorbidity indexes for head and neck cancer may be useful for prognostication of patient outcomes and predicting costs.
Patients with head and neck cancer squamous cell carcinoma are typically older than the general population and often have concurrent comorbid conditions.1,2 These comorbidities may be relatively benign, such as peptic ulcer disease or hypertension, or they may be severe and threaten survival, such as myocardial infarction, ventricular arrhythmia, and stroke.3 Many risk factors for head and neck cancer squamous cell carcinoma, such as tobacco and alcohol abuse, also place patients at risk for atherosclerosis and other malignancies.4 Severe comorbidities may drive treatment selection and affect patient outcomes. For example, patients with morbidities may not be candidates for long extirpative surgical conditions and, therefore, may be more likely to be treated by chemotherapy and/or radiation. Several studies of cancer prognosis suggest that concurrent comorbidities may significantly affect survival, even after controlling for disease stage.1
Comorbidity indexes such as the Charlson Comorbidity Index (CCI), a general comorbidity index calibrated to the general medical population, and the Washington University Head and Neck Cancer Comorbidity Index (WUHNCCI), specifically calibrated to the head and neck cancer population, can be used to improve the description of patients with head and neck cancer by better measurement and quantification of the comorbid illnesses that affect prognosis and treatment selection.5- 8 Whether these indexes are useful in predicting costs is unknown. The purpose of this research was to determine whether these indexes might also be useful for predicting cost of treatment for elderly patients with head and neck cancer.
This was a retrospective, noncontrolled, nonrandomized study. We studied 2520 Medicare patients with squamous cell cancer of the lip, oral cavity, larynx, and oropharynx who were treated between 1984 and 1994 using the Surveillance, Epidemiology, and End Results (SEER)-Medicare–linked database.9,10 We limited the analysis to patients who were treated with surgery or combined surgery and radiotherapy and included patients with local, regional, or distant disease.
The CCI was computed using the Deyo adaptation for International Classification of Diseases, Ninth Revision codes.6 The WUHNCCI was computed as described by Piccirillo et al.3
Mean cumulative 1-year and 5-year Medicare payments for all health care services were estimated for each patient. This was done by summing all Medicare payments for each patient up to 1 or 5 years or until the patient died. Note that this includes all payments for all health care services, not just for cancer care. In addition, this method does not account for censoring due to death.
Our statistical analysis was designed to fit linear models to 1-year and 5-year costs. This would allow us to study the effect of the WUHNCCI and the CCI on both short- and long-term costs. Histograms of 1-year and 5-year costs (Figure 1) suggested that the cost data violated the normality assumptions for linear regression. To account for the skewness of the data, we fit costs to generalized linear models assuming gamma-distributed errors and a natural log-link function. We estimated the impact of the comorbidity indexes on costs, controlling for demographics, primary tumor site, tumor stage (local, regional, or distant), and treatment modality. All analyses were performed using Stata SE software (version 8; StataCorp, College Station, Tex).
Patient characteristics are presented in Table 1, including demographics, primary tumor site and stage, treatments, comorbidities, and outcomes. As seen in Table 1, patients were predominantly white and male, with a mean age of 73 years. Most cancers occurred in the lip and oral cavity (50.3%), followed by the larynx (33.3%). Most disease was local (45.8%), although there were almost as many patients with regional disease (41.6%). Treatment was evenly split between surgery (48.3%) and surgery followed by radiation (51.7%). One-year survival was approximately 97.6%, dropping to 28.9% by year 5. Costs rose on average from $15 419 at 1 year to $22 299 at 5 years.
Results of our multivariate analysis of the effect of comorbidities on 1-year costs are presented in Table 2. Neither the WUHNCCI nor the CCI was significantly associated with 1-year costs. The association between the WUHNCCI and 5-year costs was statistically significant (P<.001), but the CCI was not significantly associated with 5-year costs (Table 2).
In addition, there were other factors besides comorbidity scores that were associated with costs. Stage, site, sex, race, age, and treatment were also significant predictors of costs (P<.05). Specifically, the presence of regional and distant metastases are predictive of both 1-year and 5-year costs, which may be expected since these parameters reflect patients with stage III and IV cancer. Likewise, the use of radiation is predictive of cost at both 1 and 5 years. However, the negative coefficient implies that patients treated with radiation therapy have, on average, significantly lower costs than those not treated with radiation. This is an interesting finding given that this modality is used in conjunction with surgery in patients with more advanced disease (stages III and IV). Female sex is predictive of 1-year but not 5-year costs. Race is predictive of 5-year but not 1-year costs, and age is predictive of both 1-year and 5-year costs. Laryngeal primary cancer is predictive of 1-year but not 5-year costs.
To further explore the effect of the WUHNCCI on costs, we present the marginal effect on 1-year and 5-year costs (Figure 2). The effect of the comorbidity index is nonlinear, so the effect of a 1-point increase from 1 to 2 will not be the same as a 1-point increase from 9 to 10. In presenting the marginal effects, we assumed a 60-year-old white man with local disease, treated with single modality surgery. As seen in Figure 2, a 1-point increase in the WUHNCCI (from 4 to 5) is associated with an increase in 5-year costs of approximately $2105. Furthermore, the effect of a 1-point increase in the WUHNCCI on 5-year costs depends on the starting point. An increase from 0 to 1 is associated with approximately $1658 higher 5-year costs, and an increase from 9 to 10 is associated with approximately $2837 additional costs over 5 years.
These results suggest that comorbidity indexes for head and neck cancer may be useful for prognostication of patient outcomes and predicting costs. Moreover, the more specific WUHNCCI had better predictive power in the cost models than the more general CCI, suggesting that an index of comorbidity specific to head and neck cancer may have greater utility in predicting resource use.
Several limitations need to be acknowledged. As with any study using Medicare data, all patients were 65 years or older, which raises concerns about generalizability to younger patients. This is particularly true for head and neck cancer because the average age at the time of diagnosis of oral cavity or oropharynx cancer is 56.11 There are also other concerns that arise from the use of a Medicare population. The Medicare databases do not contain pharmacy costs; therefore, costs for pharmaceuticals could not be included in the analyses. A related limitation is that patients enrolled in Medicare managed care organizations would not have any cost data. Enrollment in Medicare managed care increased over the time frame of this study, and we were not able to include these patients in any analysis of costs. A final limitation is that our methods for cumulative costs do not account for censoring because of death. Thus, these results must be viewed as conditional on the survival observed in the sample.
Prediction of costs is a powerful tool for health services researchers to study resources required to care for various diseases such as head and neck cancer squamous cell carcinoma. These studies could provide data for evidence-based analysis of various diseases and/or therapeutic interventions. These data may be used by policy makers or payers to allocate limited resources for various health conditions or interventions. The WUHNCCI appears to show promise in stratifying patients with head and neck cancer squamous cell carcinoma for the purpose of evaluating which therapeutic interventions might yield the most clinical and cost-effective outcomes.
Correspondence: Christopher S. Hollenbeak, PhD, Penn State College of Medicine, 600 Centerview Dr, MC A210, Hershey, PA 17033 (firstname.lastname@example.org).
Submitted for Publication: March 22, 2006; final revision received July 12, 2006; accepted August 17, 2006.
Author Contributions: Drs Hollenbeak and Piccirillo had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis. Study concept and design: Hollenbeak, Stack, and Piccirillo. Acquisition of data: Hollenbeak and Piccirillo. Analysis and interpretation of data: Hollenbeak, Daley, and Piccirillo. Drafting of the manuscript: Hollenbeak. Critical revision of the manuscript for important intellectual content: Hollenbeak, Stack, Daley, and Piccirillo. Statistical analysis: Hollenbeak and Daley. Obtained funding: Piccirillo. Administrative, technical, and material support: Stack and Piccirillo. Study supervision: Hollenbeak and Stack.
Financial Disclosure: None reported.
Acknowledgment: This study used the linked SEER-Medicare database. The interpretation and reporting of these data are the sole responsibility of the authors. The authors acknowledge the efforts of the Applied Research Program, National Cancer Institute; the Office of Research, Development, and Information, Centers for Medicare & Medicaid Services; Information Management Services (IMS), Inc; and the SEER Program tumor registries in the creation of the SEER-Medicare database.