Fax: (713) 563-4243
Costs of uncontrolled chemotherapy-induced nausea and vomiting among working-age cancer patients receiving highly or moderately emetogenic chemotherapy
Article first published online: 13 JUN 2007
Copyright © 2007 American Cancer Society
Volume 110, Issue 3, pages 678–685, 1 August 2007
How to Cite
Tina Shih, Y.-C., Xu, Y. and Elting, L. S. (2007), Costs of uncontrolled chemotherapy-induced nausea and vomiting among working-age cancer patients receiving highly or moderately emetogenic chemotherapy. Cancer, 110: 678–685. doi: 10.1002/cncr.22823
- Issue published online: 18 JUL 2007
- Article first published online: 13 JUN 2007
- Manuscript Accepted: 2 APR 2007
- Manuscript Revised: 22 MAR 2007
- Manuscript Received: 14 DEC 2006
- MGI Pharma Inc
- chemotherapy-induced nausea and vomiting;
- claims data;
- direct costs;
- indirect costs;
- 5-HT3 receptor antagonists
Chemotherapy-induced nausea and vomiting (CINV) is among the most feared side effects of cancer treatment. Poorly controlled CINV may lead to additional office visits or emergency room admissions, thus increasing the overall costs of cancer care. The objective of the project was to estimate the societal costs of uncontrolled CINV among working-age cancer patients.
The 1997–2002 Health and Productivity Management database, a proprietary database linking medical claims to work loss information, was used. The study population consisted of employees or their spouses who were cancer patients treated with highly or moderately emetogenic chemotherapy regimens. Costs of uncontrolled CINV were estimated by comparing the direct medical costs and indirect costs between those with and without uncontrolled CINV; all costs were normalized as monthly costs and updated to 2006 US dollars. The Wilcoxon Mann-Whitney test was used to compare the costs differences in univariate analyses, followed by multivariate analyses.
In all, 2018 patients were identified; 1771 (88%) received 5-HT3 receptor antagonists, and uncontrolled CINV was found in 563 (28%). The estimated monthly medical costs associated with uncontrolled CINV were approximately $1300 higher for cancer patients at working ages. Subgroup analysis concluded that indirect costs per patient per month were $433 higher for those in the uncontrolled CINV group.
Despite a prevalent use of the 5-HT3 receptor antagonists, uncontrolled CINV remained a common and costly problem among cancer patients treated with highly or moderately emetogenic chemotherapy. Cancer 2007. © 2007 American Cancer Society.
Chemotherapy-induced nausea and vomiting (CINV) is among the most common and feared side effects reported by cancer patients.1–3 The incidence of CINV ranges from less than 10% in patients treated with chemotherapy agents with a low emetic risk to between 30% and 90% among those whose regimens contain agents with a high emetic risk.4, 5 The introduction of 5-hydroxytryptamine-3 receptor antagonists (5-HT3-RAs), such as dolasetron (Anzemet), granisetron (Kytril), and ondansetron (Zofran), in the early 1990s made a substantial improvement in the clinical management of CINV. Thus, earlier antiemetic guidelines recommended the use of these first-generation 5-HT3-RAs for the treatment and prevention of CINV.4, 6 Further clinical improvement was achieved with 2 antiemetic agents launched in 2003—palonosetron (Aloxi), a long-acting serotonin antagonist,7–9 and aprepitant (Emend), a neurokinin-1 antagonist.10, 11 These 2 antiemetics are now included in the most recent practice guidelines.12, 13
Poorly controlled CINV was found to be associated with higher medical costs due to events such as unscheduled office visits, needs for hydration, and emergency room (ER) admissions.14–16 Most previous economic studies of CINV compare the costs or cost-effectiveness of a first-generation 5-HT3-RA with antiemetics that were considered the standard of care at the time of the study, such as prochlorperazine or metoclopramide.14, 17–19 As the first-generation 5-HT3-RAs have become the standard of care for CINV, it is important to update the economic studies to reflect the paradigm shift in the prescription of antiemetics after the mid-1990s. This update will provide more accurate cost estimates to decision-makers—estimates that take into consideration the higher acquisition costs and superior clinical performance of the 5-HT3-RAs. Additionally, these updated estimates will establish a new baseline for future economic studies comparing newer classes of antiemetics with the up-to-date standard of care for CINV (ie, 5-HT3-RAs).
This study examined the economic impact of poorly controlled CINV among working-age cancer patients who were treated with highly or moderately emetogenic chemotherapy in the late 1990s and early 2000s. Uncontrolled CINV among cancer patients in this age group may exert an economic burden on employees or their spouses. Specifically, employees with uncontrolled CINV will incur medical costs and indirect costs due to lost workdays. Additionally, employees whose spouses suffer from uncontrolled CINV will contribute to increases in medical costs if the spouses are covered by the employees' insurance. We then used a hypothetical new antiemetic medication to explore the economic implication of prophylactic antiemetic use with new agents based on the estimated costs of uncontrolled CINV.
MATERIALS AND METHODS
The primary data source used was the 1997–2002 Medstat MarketScan Health and Productivity Management (HPM) database, a large, nationwide, employment-based database collected from approximately 45 large employers in the US and over 100 health insurance payers.20 The HPM data include productivity information collected from the employers and claims data for employees and their spouse or dependents enrolled in the health care plans provided by these employers. The productivity files report information on 3 types of work leave: absence, short-term disability, and workers' compensation. The claims data contain records of inpatient and outpatient services and outpatient prescription drugs. Additionally, the HPM data include enrollment records. All files can be linked using de-identified person identifiers.
This study was approved by the University of Texas M. D. Anderson Cancer Center Institutional Review Board.
Study Population and Duration of Observation
The selected population included cancer patients who received chemotherapy between January 1 1997 and December 31 2002 whose treatment regimen contained 1 or more highly or moderately emetogenic agents. Cancer patients were identified using the ICD-9 codes and chemotherapy drugs were identified using codes from the Current Procedural Terminology (CPT) and the Healthcare Common Procedure Coding System (HCPCS). In addition, antiemetics administered as injections were identified from the HCPCS codes in the inpatient and outpatient claims, and those administered orally from the National Drug Codes (NDC) in the prescription drug claims. The level of emetogeneity for each chemotherapy agent was determined using the same classification criteria reported in the 2004 Perugia Antiemetic Guidelines.5, 21 We defined the index date as the first date that chemotherapy appeared in the claims data and applied the following inclusion criteria: 18 years and older; having a continuous enrollment or a continuous record of claims for at least 3 months; and having a duration of chemotherapy greater than 21 days.
The study duration was from the index date to the last date of chemotherapy in the claims, 6 months after the index date, or the last date available in our data (ie, December 31, 2002), whichever occurred first. The study population was categorized as patients with uncontrolled versus controlled CINV. Patients were assigned to the “uncontrolled CINV” group if they had at least 1 office visit, ER visit, or hospitalization with an ICD-9 diagnosis of nausea and/or vomiting (780.0, 780.01, 780.02, 780.03) or dehydration (276.5, 276.50) during the study period.16
Analysis of Direct Medical Costs
We compared the direct medical costs between patients with controlled and uncontrolled CINV. It is important to recognize that the observed cost differences between these 2 groups could be due to differences in chemotherapy costs or other side effects that were disproportionately observed in the CINV group. Therefore, we first categorized the total medical costs into 3 components: chemotherapy; treatments for costly chemotherapy-related adverse events such as neutropenia, or the use of granulocyte-colony stimulating factor (G-CSF) while excluding CINV; and other interventions. We then used the cost difference in the “others” category between groups to infer the direct medical costs associated with uncontrolled CINV. This approach has the advantage of minimizing biases in cost estimation due to unequal costs of chemotherapy agents or side effects in the 2 groups.
Because medical costs were highly skewed, Wilcoxon Mann-Whitney tests were performed to compare the medical costs between groups in the univariate analysis, and the logarithm transformation of costs were used as the dependent variable in the multivariate analyses. Covariates in the multivariate analyses included patients' demographics, geographic characteristics, relationship with the employers (employees or spouses of employees), clinical factors, and time trends. Clinical factors included primary cancer types, classified as breast, lung, gastrointestinal (GI), lymphoma, and other types, comorbidities,22, 23 and tumor metastasis status. We obtained the cost difference between patients in the controlled and uncontrolled CINV groups by applying appropriate econometric technique (known as the smearing estimate) to retransform the difference in log-transformed cost estimated in the regression model to the original scale.24, 25 All costs were updated to 2006 dollars using the medical care component of the consumer price index. Results are reported as cost per patient per month by first aggregating costs from all relevant medical claims throughout the study period and then normalizing the aggregated amount to monthly costs. Monthly costs were reported as this measure approximates a cycle of chemotherapy.
Analysis of Indirect Costs (Exploratory Subgroup Analysis)
The HPM data provide information of absence leave for a subset of employees. As an exploratory analysis, we used these employees to assess the indirect costs of uncontrolled CINV; employees with short-term disability claims during the entire study period were excluded. We estimated the indirect costs by comparing the work-loss days between employees in the uncontrolled and controlled CINV groups. We imputed the indirect costs by multiplying hours lost from work by the average hourly wage rate. Wage rates were obtained from the Bureau of Labor Statistics and stratified by census regions to account for geographic variations in wage structure.26 We inflated the wage rates to 2006 dollars using the consumer price index for urban wage earners and clerical workers.27 Analysis of indirect costs followed a similar analytical plan as in the direct medical costs. All P-values reported are 2-sided; statistical significance is defined as P < .05.
Analysis of Economic Implication of Prophylactic Use of Newer Antiemetics
Upon obtaining the estimated average per-patient costs associated with uncontrolled CINV, we conducted a number of sensitivity analyses with a hypothetical new antiemetic medication to explore whether prophylactic antiemetic use with the newer agents can be justified economically. To determine the economic implication of the hypothetical new agent, we considered 3 factors in addition to the average cost of uncontrolled CINV: the rate of uncontrolled CINV (RCINV) before the launch of this hypothetical new agent, the effectiveness (E) of the new agent in reducing such rate, and the price (P) associated with a complete course of prophylactic antiemetic administration with the new agent. We denoted the costs of uncontrolled CINV as C and the total number of patients treated with highly or moderately emetogenic chemotherapy as N, prophylactic use of this hypothetical new antiemetic would add an additional (N × P) dollars to the health care system, but would reduce the rate of uncontrolled CINV from RCINV to (RCINV × E). Therefore, if the savings achieved from reducing the costs of uncontrolled CINV outweighed the additional costs of this hypothetical new antiemetic, prophylactic use with this new agent in clinical practice would be considered economically justified. We performed threshold analyses to calculate the price of a complete course of prophylactic use with the hypothetical new antiemetic at which the additional costs of prophylactic use equaled the total savings from reduced costs of uncontrolled CINV (ie, N × RCINV × E × C) and named this price (denoted as P*) the “threshold price.” Using the hypothetical new agent as prophylactic antiemetics could be justified economically if the price of this new agent per course was lower than the above calculated threshold price (ie, P < P*). We then conducted sensitivity analyses under various scenarios of prices, rates of uncontrolled CINV, and clinical effectiveness to explore the economic feasibility of such practice pattern.
In all, 2018 working-age cancer patients treated with highly or moderately emetogenic chemotherapy were identified. Descriptive statistics in Table 1 showed that despite the prevalent use of the 5-HT3-RAs (approximately 87%), 563 out of 2018 patients (27.9%) experienced uncontrolled CINV. Table 1 also shows significant differences in the geographic distribution (P = .0013) and cancer types (P < .0001) between the controlled and uncontrolled CINV groups and a significantly higher proportion of 5-HT3-RAs use in the uncontrolled group (92.5% vs 85.9%, P < .0001).
|All patients N = 2018||Uncontrolled CINV N = 563||Controlled CINV N = 1455||P|
|No. (%)||No. (%)||No. (%)|
|Age, y||50.62 (8.14)||50.48 (7.94)||50.68 (8.22)||.6334|
|Men||621 (30.77)||186 (33.04)||435 (29.90)||.1704|
|Women||1397 (69.23)||377 (66.96)||1020 (70.10)|
|Employees||1117 (55.35)||331 (58.79)||786 (54.02)||.0531|
|Spouses||901 (44.65)||232 (41.21)||669 (45.98)|
|Northeast||498 (24.68)||114 (20.25)||384 (26.39)||.0013|
|North Central||593 (29.39)||152 (27.00)||441 (30.31)|
|South||788 (39.05)||253 (44.94)||535 (36.77)|
|West||139 (6.89)||44 (7.82)||95 (6.53)|
|Breast||903 (44.75)||218 (38.72)||685 (47.08)||<.0001|
|Lung||387 (19.18)||130 (23.09)||257 (17.646)|
|GI||234 (11.60)||88 (15.63)||146 (10.03)|
|Lymphoma||187 (9.27)||53 (9.41)||134 (9.21)|
|Others||307 (15.21)||74 (13.14)||233 (16.01)|
|High||260 (12.88)||85 (15.10)||175 (12.03)||.0648|
|Moderate||1758 (87.12)||478 (84.90)||1280 (87.97)|
|Use of 5-HT3||1771 (87.76)||521 (92.54)||1250 (85.91)||<.0001|
|1997||174 (8.62)||45 (7.99)||129 (8.87)||.0008|
|1998||223 (11.05)||65 (11.55)||158 (10.86)|
|1999||313 (15.51)||119 (21.14)||194 (13.33)|
|2000||391 (19.38)||100 (17.76)||291 (20.00)|
|2001||494 (24.48)||133 (23.62)||361 (24.81)|
|2002||423 (20.96)||101 (17.94)||322 (22.13)|
Analysis of Direct Medical Costs
Our univariate analysis showed that the total direct medical costs per patient per month for the uncontrolled CINV group were $1797 (= $10,720 − $8923) higher than that for the controlled CINV group (P < .0001) (Table 2). However, the cost decompositions suggest that not all the differences can be attributed to uncontrolled CINV. After excluding the costs of chemotherapy and its related adverse events, the difference in monthly costs reduced to $1383 (P < .0001).
|Cost Categories||Uncontrolled CINV||Controlled CINV||P|
|Total costs||$10,720 ($7829)||$8923 ($7459)||<0.0001|
|Chemotherapy||$4685 ($3939)||$4333 ($4448)||.0003|
|Treatments of chemo-related adverse events|
|Neutropenia||$389 ($1241)||$345 ($1477)||.0006|
|Blood products||$7 ($134)||$1 ($18)||.7280|
|G-CSF||$89 ($710)||$76 ($419)||.0479|
|Other costs||$5551 ($5441)||$4168 ($4800)||<.0001|
Table 3 presents the results from the regression analysis with log(cost) as the dependent variable. For ease of interpretation, we converted the regression coefficients to percentages28; for example, it shows that on average the monthly medical cost of patients in the uncontrolled CINV group was 29.79% (= 100* (exp(0.261) − 1)) higher than that of patients in the controlled CINV group, after controlling for confounding factors such as demographic, socioeconomic, and clinical characteristics. After retransformation, this magnitude of difference in log-transformed cost was equivalent to approximately $1280.
|Coefficients||% Difference in Costs||P|
|Male (vs female)||0.068||7.07%||.255|
|Employee (vs spouse)||0.057||5.82%||.222|
|Highly emetic (vs moderately)||0.332||39.38%||<.001|
|Cancer type (reference group = breast cancer)|
|Region (reference group = northeast region)|
|Year (reference group = 1997)|
Exploratory Subgroup Analysis
In all, 94 employees met our inclusion criterion for analyses of indirect costs described above; 30 of the 94 (31.9%) patients had uncontrolled CINV. The average number of work-loss days for uncontrolled and controlled CINV groups was 6.23 and 3.61 days per month (P = .120). We also found that employees in the uncontrolled CINV group were more likely to take a leave of absence, although the difference was not statistically significant (60.0% vs 45.3%, P = .184). After adjusting for demographic and clinical characteristics, we found that on average the total, direct, and indirect costs of patients experiencing uncontrolled CINV were $1832 (P = .047), $1399 (P = .126), and $433 (P = .102) higher than those who did not experience uncontrolled CINV. Note that P-values in this sample should be interpreted with caution, as the sample size is underpowered to detect statistically significant differences between patients in the controlled and uncontrolled CINV groups.
Economic Feasibility of Prophylactic Antiemetic Use With a New Agent
Using the adjusted costs of uncontrolled CINV (ie, $1280) estimated above, our sensitivity analyses varied the rate of uncontrolled CINV from 0% to 90% at 10% intervals and considered 4 levels of effectiveness, ranging from a 20% reduction at the rate of uncontrolled CINV to 100%. The threshold price corresponding to each level of uncontrolled CINV in the absence of our hypothetical new agent in combination of the clinical effectiveness of the agent is presented in Figure 1; these prices represented the maximum dollar value of a complete course of prophylactic use with the hypothetical new antiemetic that can be sustained by the health care system economically under various scenarios. For example, if the rate of uncontrolled CINV before the launch of the hypothetical new antiemetic was 30%, and the new agent was expected to reduce the rate by 80%, then prophylactic use of this antiemetic was economically beneficial if the price of a complete course of prophylactic antiemetic use was lower than $307. Figure 1 shows a clear trend of increasing threshold prices as the rate of uncontrolled CINV or the effectiveness of the hypothetical new agent increases.
This study estimated costs associated with uncontrolled CINV among working-age cancer patients in an era when the use of the first-generation 5-HT3-RAs has become the standard of care for the prevention and treatment of CINV. Despite this major improvement in supportive care, there is still much room for improvement. The estimated additional direct medical costs per patient per month due to uncontrolled CINV was close to $1300, or 30% higher than those in the controlled CINV group.
Despite a high compliance with the antiemetic guidelines that recommend the use of 5-HT3-RAs, several studies reported that cancer patients still experienced CINV and that the problem seemed to be worse in delayed CINV.29–31 Grunberg et al.31 reported that 13% and 37% of cancer patients receiving moderately emetogenic chemotherapy experienced acute nausea and vomiting, respectively. For patients experiencing delayed CINV, the percentages increased to 28% and 52%, respectively. The rate of delayed CINV is even higher among patients whose chemotherapy regimens contain 1 or more highly emetogenic agent, such as cisplatin. Kris et al.32 reported a 38% and 61% rate of acute and delayed CINV, respectively, among patients treated with cisplatin and received metoclopramide and dexamethasone at the time of chemotherapy administration. Vanscoy et al.16 found that among cancer patients treated with moderately emetogenic chemotherapy and given prophylactic 5-HT3-RAs, 11.1% reported extreme CINV events, defined as having 1 or more unscheduled visits or calls to the oncology practice for management of failed CINV prophylaxis. Our findings are comparable to previous research; we found that, whereas close to 90% of our study sample received 5-HT3-RAs, 28% still experienced uncontrolled CINV.
Our analysis was based on claims data from 1997 to 2002, a time during which the newer antiemetics such as palonosetron and aprepitant were not yet available. It is possible that some of the higher acquisition costs of these new agents can be offset by cost savings achieved from reducing the rate of uncontrolled CINV, especially from delayed CINV. The threshold prices calculated in our sensitivity analyses using a hypothetical new agent provide some insights as to the economic implications of these new agents that may inform clinical choices. If the rate of uncontrolled CINV before the administration of any new agent is 30% and that the new agent can reduce the rate of uncontrolled CINV by 80% (ie, from 30% to 6%), our study suggests that an antiemetic regimen that increases the antiemetic costs by $307 or less is likely to reduce the overall costs. Or, similarly, if the rate is 50% and the new antiemetic can reduce it to 10% (ie, 80% effectiveness), our study suggests that an antiemetic regimen that increases the costs by $512 or less would save money.
We were able to obtain absence records for a small group of employees in our data. To explore whether this subgroup of patients and the full sample have similar characteristics, we compared patient characteristics between these 2 samples and found that these 2 samples were similar in general, with exceptions in the distribution of gender, geographic location, and time trend. The higher proportion of patients in the subgroup observed in year 2000 and 2001 could probably explain the slightly higher cost difference in direct medical costs found in the subgroup ($1399 vs $1280), as a positive association between medical costs and time trend was indicated in the full sample analysis (Table 3). After adjusting for demographic and clinical characteristics, our exploratory analysis concluded that productivity loss ($433) accounted for approximately 24% of the total cost difference ($1832) between the uncontrolled and controlled CINV groups. Assessing productivity loss for conditions such as uncontrolled CINV that arise from a major medical treatment is challenging. First, one cannot evaluate productivity loss for uncontrolled CINV for patients who are on short-term disability leave because the absence records are not meaningful for such employees. Even cancer patients who stay active in the labor force may exhaust all their work leave due to frequent medical office visits for confirmatory diagnoses, evaluations, and chemotherapy. It is possible that employers allow these employees to work at a lesser capacity (eg, switching to part-time), or simply to take days off when necessary and make up for the hours later. Either scenario will fail to generate a recorded day of absence from work, even though productivity losses had actually occurred. The HPM data do not provide the information that would allow us to differentiate various scenarios; consequently, the indirect costs reported in our study may be underestimated.
This study focused on working-age cancer patients because cancer in young adult patients is often treated more aggressively; therefore, the likelihood of observing a higher cost of uncontrolled CINV is greater in young adults. Additionally, the selection of this population allowed us to have more comprehensive evaluations of the economic burden, by including direct and indirect costs. Whether the results of our study can be generalized to elderly cancer patients is yet to be determined. Factors such as a smaller proportion of patients treated with chemotherapy or a larger proportion of patients receiving agents with low emetic properties may lead to a lower cost of uncontrolled CINV in the elderly population than that estimated from our study. On the other hand, uncontrolled CINV may have more severe health consequences among elderly cancer patients and thus result in higher costs in this population.
Our study was constrained by information collected in claims data. For example, we were not able to differentiate between uncontrolled, acute and delayed CINV. The distinction between the 2 types of emesis would have provided additional information on the source of the economic benefit one can expect from the newer antiemetics. A recent meta-analysis concluded that the use of the first-generation 5-HT3-RAs beyond 24 hours after chemotherapy for prevention of delayed emesis was not supported by clinical or economic evidence.33 Because the newer antiemetics have demonstrated superior clinical performance in reducing delayed emesis, some of the higher acquisition costs of these drugs may be offset by cost savings achieved from lowering the rate of uncontrolled, delayed CINV. Consequently, economic evidence may support using the newer agents to prevent delayed emesis. Unfortunately, limited clinical information available in claims data precluded such comparison. Furthermore, our estimates of indirect cost and total costs were derived from a small subset of patients whose absence records were available. Due to the limited power provided in this sample, results from the subgroup analysis of indirect costs should be viewed as exploratory and should be confirmed with a larger sample in future studies.
Sponsored by a research grant from MGI Pharma Inc.
- 7Improved prevention of moderately emetogenic chemotherapy-induced nausea and vomiting with palonosetron, a pharmacologically novel 5-HT3 receptor antagonist: results of a phase III, single-dose trial versus dolasetron. Cancer. 2003; 98: 2473–2482., , , et al.
- 8Palonosetron improves prevention of chemotherapy-induced nausea and vomiting following moderately emetogenic chemotherapy: results of a double-blind randomized phase III trial comparing single doses of palonosetron with ondansetron. Ann Oncol. 2003; 14: 1570–1577., , , et al.
- 10The oral neurokinin-1 antagonist aprepitant for the prevention of chemotherapy-induced nausea and vomiting: a multinational, randomized, double-blind, placebo-controlled trial in patients receiving high-dose cisplatin–the Aprepitant Protocol 052 Study Group. J Clin Oncol. 2003; 21: 4112–4119., , , et al.
- 11Addition of the neurokinin 1 receptor antagonist aprepitant to standard antiemetic therapy improves control of chemotherapy-induced nausea and vomiting. Results from a randomized, double-blind, placebo-controlled trial in Latin America. Cancer. 2003; 97: 3090–3098., , , et al.
- 12National Comprehensive Cancer Network. Antiemesis: clinical practice guidelines in oncology v1. 2005. Available at URL: http://www.nccn.org/professionals/physician_gls/PDF/antiemesis.pdf Accessed on January 2, 2006.
- 13The Antiemetic Subcommittee of the Multinational Association of Supportive Care in Cancer (MASCC). Prevention of chemotherapy- and radiotherapy-induced emesis: results of the 2004 Perugia International Antiemetic Consensus Conference. Ann Oncol. 2006; 17: 20–28., , .
- 16Preventing chemotherapy-induced nausea and vomiting: the economic implications of choosing antiemetics. Commun Oncol. 2005; 2: 127–132., , , .
- 20Thomas Medstat. 1997–2002 MarketScan Health and Productivity Management Database User Guide and Data Dictionary. Ann Arbor, MI: Thomas Medstat; 2003.
- 21MASCC. Antiemetic Guidelines, vol. 2007. Perugia, 2005. Available at URL: http://www.mascc.org/media/Resource_centers/MASCC_Guidelines_Update_9_05.pdf Accessed on February 27, 2007.
- 26Bureau of Labor Statistics. National Compensation Survey: Occupational Wages in the United States, 1997–2002. Available at URL: http://www.bls.gov/ncs/ncspubs.htm#Compensation Accessed on December 22, 2005.
- 27Bureau of Labor Statistics. Consumer Price Index — Urban Wage Earners and Clerical Workers. Available at URL: http://data.bls.gov/cgi-bin/surveymost Accessed on December 22, 2005.
- 28The interpretation of dummy variables in semilogarithmic equations. Am Econ Rev. 1980; 70: 474–475., .