Laryngeal Cancer Cost Analysis: Association of Case-Mix and Treatment Characteristics With Medical Charges^†

INTRODUCTION

Cancer of the larynx is the most common cancer of the upper aerodigestive tract.¹ The evaluation and management of laryngeal cancer consumes a large fraction of the health care resources used in the management of all upper aerodigestive tract cancers. To better understand the factors associated with the cost of managing laryngeal cancer, we have undertaken a cost-identification analysis of laryngeal cancer patients managed at the University of Iowa Hospitals and Clinics (UIHC) between January 1, 1991 and December 31, 1994.

Within health care there are essentially four types of cost analyses performed: 1) cost identification, 2) cost-benefit, 3) cost-effectiveness, and 4) cost-utility.² Cost-identification analysis is a simple accounting of the cost incurred as the result of an intervention or program and does not incorporate measures of health benefit. In cost-benefit analysis, all costs and health benefits are assigned monetary values, and an overall net gain or loss can be calculated solely in financial terms. In cost-effectiveness studies, the monetary cost of an intervention or program is evaluated in concert with an identified unit of health care output (e.g., cost per mm Hg of lowering diastolic blood pressure on one antihypertensive medication vs. another). In cost-utility studies, the structure is similar to cost-effectiveness studies but the measure of health care output is represented by a more general unit such as the quality-adjusted life year (QALY).^3-5

A crucial consideration when conducting this type of analysis is the definition of cost. The economic definition of cost is calculated as the value of the consumed resources if they had been put to use for an alternative purpose.^{2, 6, 7} If one uses this definition, it is obvious that hospital charges do not necessarily represent true hospital costs. Unfortunately, the true hospital costs associated with an intervention are generally very difficult to identify and hospital charges are often used as a best-approximation for hospital costs.⁶

Another fundamental consideration in performing a cost analysis is the comprehensive accounting for all resources consumed or gained as the result of the medical intervention under study. Direct costs include both direct medical costs (e.g., operating room, hospital room, physician, laboratory, and radiology charges) and direct nonmedical costs (e.g., transportation to the hospital, lost wages of a caregiver, and childcare expenses during parental hospitalization). There are also less easily quantified charges that remain to be considered. These include indirect costs (morbidity and mortality costs) and the intangible costs assigned to health states such as pain and disfigurement.

While accounting for resources consumed, cost analyses may also include measurement of intangible direct and indirect benefits. Direct benefits include resources gained because of the cost savings of a particular intervention when compared with an alternative intervention. Indirect benefits include primarily factors associated with patient return to economic productivity. Intangible benefits include factors related to the value placed on improved health. Quantifying such factors as future health benefits or costs that result from the current intervention may also be considered, but these analyses can become quite complex and require significant estimation.^{2, 4, 6}

In performing cost analyses, it is important to consider the perspective from which the analysis is being conducted. Cost as realized by the hospital is very different from the true costs borne by a third-party payer or the patient, and this is different from what the cost would be if the perspective of society is taken. For the purposes of this study, the perspective is that of the payer. This matter of perspective is paramount because the manner in which each of these parties realizes the involved costs directly influences the economic impact of any given intervention.^{2, 6}

In this analysis, a cost-identification strategy was used to evaluate the relationships between direct medical charges for laryngeal cancer management and patient-mix variables, including comorbid illness and treatment modality, during defined billing periods representing pretreatment evaluation, treatment, and follow-up. This type of analysis was selected because it is suited to evaluate the associations between patient characteristics, treatment modalities, and charges generated in each of the selected time periods. The goals of this study were to identify patient-mix or treatment variables that are associated with increased or decreased charges for each billing period. Although it is recognized that the monetary data obtained are limited in scope, this information may be useful in identifying strategies for potential cost reduction and improved resource allocation in managing this disease. In an effort to avoid the complexities of introducing a number of cost and benefit estimations into this cost-identification study, only direct medical charges were considered and no analysis of direct nonmedical, indirect, intangible, or future costs or benefits was considered.

Hospital and physician charges were used as a proxy for the ideal monetary data, which in this type of analysis would be the opportunity cost to the caregiver for providing the medical service. From a payer perspective, it might be argued that the dollar amount of the bill represents a true opportunity cost to the payer at the time the bill is paid. However, because of variable reimbursement schemes, this is not likely to be a valid argument in most cases. Additionally, the hospital charges are not representative of the true direct medical costs to the provider or society. To reflect accurately this approximation to the true costs of interest, we will refer to monetary values used in this paper as charges rather than costs.

METHODS

This study analyzed the management charges of patients diagnosed with laryngeal carcinoma at the UIHC between January 1, 1991 and December 31, 1994. Patients who were diagnosed with a recurrent cancer or who had been treated for a previous head and neck cancer were not included. Patients who were diagnosed at the UIHC but who did not receive treatment at this facility were also omitted; this included patients who generated pretreatment charges at the UIHC but who had single-modality radiation therapy at another facility.

Patients diagnosed with laryngeal carcinoma between 1991 to 1994 were initially identified through the hospital's cancer registry. Medical chart audits were then performed to determine which of these patients met the eligibility criteria relating to history of head and neck cancer and location of treatment. A total of 57 patients were identified for inclusion in the study.

Case-mix and treatment variables were collected for each patient during the chart-review process. These elements included age at diagnosis, sex, comorbid illness present at the time of diagnosis, packs/year tobacco use, location of the tumor within the larynx, clinical TNM classifications, treatment modalities, and outcome (recurrence of disease and death).

Based on the documented coexisting diseases at the time of diagnosis, each patient was assigned a Kaplan-Feinstein cogent comorbidity grade.⁸ This grade is based on the severity of illness in numerous categories including hypertension, cardiac, cerebral or psychic, respiratory, renal, hepatic, gastrointestinal, peripheral vascular, malignancy, locomotor, and alcoholism. Using a four-point scale, comorbidity is classified as not present (0), mild (1), moderate (2), or severe (3). Measurements of the patients' physical status included the ASA class. The ASA class is a score assigned by anesthesiologists using a five-point scale as follows: no organic, physiological, biochemical or psychiatric disturbance (class 1), mild to moderate systemic disturbance (class 2), severe systemic disturbance (class 3), life-threatening disturbance (class 4), or moribund patient with little chance of survival (class 5).⁹

The billing records for these patients were obtained from the hospital's billing department. Total medical charges were calculated for four different periods: pretreatment (from diagnosis to treatment), 0 to 3 months, 3 to 12 months, and 12 to 24 months. “Zero-time” was the initiation of cancer-directed treatment. If patients received treatment at the time of diagnosis (e.g., laser excision of small glottic cancer), all charges were included within the 0- to 3-month period.

These charges represent all physician, radiographic, laboratory, office visit, inpatient, and operating room charges incurred at the UIHC, including any charges not directly related to treating the laryngeal cancer (e.g., pretreatment consultations with other services required to proceed with treatment of the laryngeal cancer). These charges did not include any charges incurred outside the UIHC. Based on the chart reviews, very little medical care was received by the patients outside the UIHC during the billing periods under consideration. The most frequent medical care received outside the UIHC during the treatment and follow-up periods was an occasional office visit to a referring physician or general practitioner in a patient's hometown.

To normalize the charges incurred across these 4 years, all charges were converted to 1983 dollars using Citibase's medical care consumer price index. This was the last year for which the consumer price index was normalized at the time of our analysis.¹⁰ Each patient's charges were normalized using the average monthly index factor during which their charges were incurred. The average monthly index factor for the first 12-month period was used to convert the pretreatment, 0- to 3-, and 3- to 12-month charges, and the average monthly index factor for the second 12-month period was used to convert the 12- to 24-month charges. If patients died or were lost to follow-up during a given period, whatever charges they incurred during that time were included. For subsequent periods, patients who were lost to follow-up were excluded from analysis, whereas patients who died were included in the analysis.

Statistical analyses were performed using SAS.¹¹ Because the financial data were skewed by a few patients' having charges substantially larger than the average, an initial step involved converting these nonparametric data logarithmically to perform parametric analyses. The next step involved univariate analyses (using ANOVAs, t tests, and linear regressions) to identify which patient, disease, and treatment characteristics were associated with the medical charges within each of the separate billing periods.

The final step involved multivariate analyses to determine how well these variables accounted for the variability within the medical charges for each separate billing period, and the total 1- and 2-year billing periods. The stepwise regression approach, which evaluated independent variables for entry into and removal from the model, was used to build a regression model in a sequential fashion. This approach allowed us to examine combinations of independent case-mix and treatment variables to explain the dependent cost variables. It was chosen over the backward-selection procedure to avoid working with more independent variables than were necessary. All variables with a P value of .15 or less in the univariate analyses were chosen for inclusion in the multivariate analysis for that specific billing period.

RESULTS

This study included 57 patients diagnosed with laryngeal cancer at this institution between January 1, 1991 and December 31, 1994. The majority were 50 to 69 years old (59.6%) and male (77.2%) (Table I). The level of comorbidity, as measured by the Kaplan-Feinstein grade, indicated that 60.7% of patients had some degree of comorbidity (32.1% being mild, 14.3% being moderate, and 14.3% being severe).

Table TABLE I.. Pretreatment Patient Characteristics (N = 57).

Disease characteristics indicated that 47.4% of these patients' tumors originated in the glottis and 52.6% in the supraglottis (Table II). The extent of tumor was fairly evenly distributed between T1 (29.8%), T2 (21.1%), T3 (29.8%), and T4 (17.5%). Most disease was localized (71.9% being N0), while 15.8% were N1, 10.5% were N2, and 1.8% were N3. The distribution of stage group indicated that the majority (61.4%) of patients had advanced disease (stage III or IV).

Table TABLE II.. Pretreatment Disease Characteristics (N = 57).

Most patients received single-modality therapy as initial treatment, with 29.8% receiving surgery and 29.8% receiving radiation therapy (Table III). Combined surgery and radiation therapy as initial treatment was performed on 36.8% of these patients. A very small percentage received radiation therapy combined with chemotherapy (1.8%) or all three modalities (1.8%) as the initial intervention.

Table TABLE III.. Treatment Modalities and Outcome (N = 57).

The outcome data presented in Table III indicate that 80% of these patients were free of disease at the time of last follow-up. Ten percent recurred locally, 4% had regional recurrence (with local recurrence), 8% had a distant recurrence (with or without local recurrence), and 2% were never disease free. At the time of last contact, 91.2% of these patients with laryngeal cancer were alive (82.4% without disease and 8.8% with disease) (Table III). Of the five patients who died, two died with cancer, two died without cancer, and one died with cancer status unknown.

The median charges (normalized to 1983 dollars) were $3,234 for the pretreatment period, $19,101 for the 0- to 3-month period, $1,080 for the 3- to 12-month period, and $549 for the 12- to 24-month period. The charges generated during the 0- to 3-month period, primarily representing initial treatment charges, accounted for the bulk of the total 1-year charge (81.6%) as well as the total 2-year charge (79.7%).

Univariate Analyses

Univariate analyses were performed to determine the relationship of various patient, disease, and treatment characteristics with charges generated during the defined billing periods. (P values are shown in Table IV and median costs of significant variables are shown in Table V.) No case-mix or treatment variables were associated with pretreatment charges. Site, staging variables, and type of treatment were associated (P < .05) with 0- to 3-month, total 1-year, and total 2-year charges. Patients with tumors originating in the supraglottis, with larger tumors (T3-4), with nodal involvement (N1-3), and with advanced-stage grouping (stage III-IV) had higher charges during these three periods. Initial treatment with radiation therapy alone was associated with lower charges ($7,483) when compared with initial treatment with surgery alone ($20,635) for the 0- to 3-month period. The difference in charges between surgery and radiation as initial treatment was found only within the 0- to 3-month billing period. For the total 1-year and 2-year periods, the initial difference between surgery alone versus radiation therapy alone was no longer present. Radiation therapy was associated with lower charges when compared with surgery combined with radiation therapy for the 0- to 3-month, and total 1-year and 2-year periods. Single-modality therapy was associated with lower charges when compared with multimodality therapy for all three of these periods. Of the patient characteristics, only ASA class was associated with total 2-year costs (P = .0325, with classes 0 to 2 being $25,588 and 3 to 4 being $35,894). Charges generated during the 3- to 12-month period and the 1-year to 2-year period indicated fewer associations. In the 3- to 12-month period, patients with nodal disease, with advanced disease, and with single-modality initial treatment had higher charges. In the 1- to 2-year period, patients with nodal disease had higher charges.

Table TABLE IV.. P Values for the Univariate Associations of Case-Mix, Treatment, and Outcome Characteristics With Charges Generated During Each Billing Period.

Table TABLE V.. Median Costs (in 1983 Dollars) for Variables That had P Values < .15 in the Univariate Analyses and Were Used in the Multivariate Analyses.

It should be noted that the primary reason for the univariate analyses was to select independent variables for inclusion in the multivariate regression analyses. In this regard, the P values for the univariate analyses are more of a tool than an end result.

Multivariate Analyses

Table VI presents the results of the stepwise, multiple regression analyses to determine the independent association of these case-mix and treatment characteristics with charges generated. (The number of resultant variables that are found to be significant is limited in multiple regression analyses by the overall numbers within each category being included in the model. Because our overall count was limited to 57 patients, it is possible that all variables significantly associated with cost were not identified.) No multiple regression analysis was performed for the pretreatment period because no variables were associated with charges generated before treatment in the univariate analysis.

Table TABLE VI.. P Values for the Multiple Regression Analysis of the Associations of Case-Mix, Treatment, and Outcome Characteristics With Charges Generated During Each Billing Period.

For the 0- to 3-month period, the presence of comorbidity (Kaplan-Feinstein grade 1-3), advanced TNM stage, and the initial use of multiple treatment modalities were independently associated with higher charges, with the strength of the model (R²) being 0.5575. For the 3- to 12-month period, nodal involvement (N1-3), advanced TNM stage group, and the use of multiple treatment modalities were independently associated with higher charges, with an R² of 0.2836. Higher 1-year total charges were associated with larger tumor size (T3-4), nodal involvement (N1-3), and the initial use of multiple treatment modalities, with an R² of 0.5246. Although several variables showed simple associations with the 1- to 2-year treatment charges, none were independently associated with the charges during this billing period when subjected to a multiple regression analysis. In the period covering the first 2 years after diagnosis, higher charges were associated with the presence of comorbidity (Kaplan-Feinstein grade 1-3) and advanced TNM stage group. This model produced an R² of 0.5055.

DISCUSSION

This study was structured as a cost-identification analysis of patients with laryngeal cancer to investigate the associations between a variety of patient-mix and treatment variables and the direct medical charges incurred in the management of this disease through a 2-year follow-up period. The project aimed to identify variables that may be associated with higher management charges and to identify strategies that could potentially reduce these management charges.

The largest proportion of the management charges were incurred during the treatment (0-3 mo) period. Pretreatment charges represented only a small percentage (13.5%) of the total 2-year management charges. While any cost reduction is beneficial, these data would suggest that dramatic reductions in management charges would not be realized through manipulation of the pretreatment work-up algorithm.

From a public health standpoint, the most effective means of decreasing the cost of laryngeal cancer management in the United States would be to eliminate the use of tobacco products. Although significant effort is being applied toward that end, this goal is very unlikely to be attained in the foreseeable future. Stage of disease and treatment modality (single vs. multimodality) were picked up in the multiple regression models for both 1-year and 2-year total charges (Table VI), with early-stage disease and single-modality management being associated with lower management charges. The relationship between early detection or cancer screening programs and improved outcome for upper aerodigestive tract cancers is likely very complex.¹² However, these results support continued investigation of measures that would allow initiation of treatment for laryngeal cancer at the earliest time after development.

Single-modality treatment was shown to be a significant predictor of lower management charges from the initiation of cancer-directed treatment through the 1-year total. By 2 years this difference was no longer present and initial treatment modality was not significant in the multivariate model predicting 2-year total costs. It is likely that charges incurred in the treatment of recurrences and management of comorbid illness throughout the 2-year follow-up contributed to the elimination of the incremental difference in treatment charges initially seen between single- and multi-modality treatment groups. Given comparable efficacy, it is tempting to favor one method of treatment over another based on the apparent up-front costs when formulating treatment algorithms. This finding points out that a variety of factors may act to eliminate the incremental cost differences that are initially present between two treatment strategies. The overall cost of a treatment strategy is composed of not only the initial costs, but also the incurred health care costs after the initial treatment and this must be considered when formulating a treatment plan based on cost-evaluation studies.

The diagnosis of laryngeal cancer covers an extremely broad spectrum of disease severity. Stratification of laryngeal cancer patients by disease severity that includes comorbid illness has been shown to improve the accuracy of disease classification with regard to predicted survival.^{13, 14} Indeed, it has been proposed that classification of head and neck cancer disease severity incorporate other considerations beyond simple anatomic index disease stage.¹⁵ The government introduced the diagnosis-related group (DRG) system for Medicare reimbursement in 1983, and many third-party payers have adopted this type of system for reimbursement. Although the Medicare DRG reimbursement coding system has undergone multiple revisions since its inception, there is no allowance within the current DRG system for stratification of laryngeal cancer patients based on stage of index disease or the presence of comorbid illness.

The results of this study indicate that reimbursement schemes that more precisely stratify patients by severity of disease would result in a more accurate use of the health care resources available for the management of laryngeal cancer. The diagnosis of laryngeal cancer encompasses a broad range of potential disease states that have a large range of management charges. Practitioners and institutions that function in a tertiary care role (managing the more advanced cases and patients with greater comorbid illness) are likely to be financially penalized under the current Medicare DRG system, and this has been demonstrated to be the case for hospital reimbursement.^16-19