Humanistic and economic impacts of hepatitis C infection in the United States

Abstract

Objective:

Prior research examining the effect of hepatitis C virus (HCV) on health-related quality of life (HRQoL) and healthcare costs is flawed because non-patient controls were not adequately comparable to HCV patients. The current study uses a propensity score matching methodology to address the following research question: is the presence of diagnosed hepatitis C (HCV) associated with poorer health-related quality of life (HRQoL) and greater healthcare resource use?

Methods:

Using data from the 2009 US National Health and Wellness Survey, patients who reported a HCV diagnosis (n = 695) were compared to propensity-matched controls (n = 695) on measures of HRQoL and healthcare resource use. All analyses applied sampling weights to project to the US population.

Results:

HCV patients reported significantly lower levels of HRQoL relative to the matched-control group, including the physical component score (39.6 vs. 42.7, p < 0.0001) and health utilities (0.63 vs. 0.66, p < 0.0001). The number of emergency room visits (0.59 vs. 0.44, p < 0.05) and physician visits (7.7 vs. 5.9, p < 0.05) in the past 6 months were significantly higher for the HCV group relative to matched controls.

Conclusion:

The results of this study suggest that HCV represents a substantial burden on patients by having a significant and clinically-relevant impact on key dimensions of HRQoL as well as on utilization of healthcare resources, the latter of which would result in increased direct medical costs.

Limitations:

Due to limitations of the internet survey approach (e.g., inability to confirm HCV diagnosis), future research is needed to confirm these findings.

Key words::

• Emergency room visits
• Health status
• Healthcare provider visits
• Hepatitis C virus
• Hospitalizations

Introduction

Hepatitis C virus (HCV) is the most common chronic blood-borne infection in the United States, affecting at least 2.7 million Americans1. Although the significant risk for the development of cirrhosis and hepatocellular carcinoma in the later stages of infection is well recognized2, clinicians commonly consider the early phases of chronic HCV infection, when patients maintain near-normal liver enzymes and histology, to be asymptomatic. However, a number of studies have shown the deleterious effect of the presence of HCV on quality of life2–5, especially in patients with advanced fibrosis6 and human immunodeficiency virus infection.

Conventionally, the priority of HCV management has been to limit disease progression and related disability and mortality7. While dozens of direct antiviral agents are under development, the current standard of care consists of pegylated interferon with ribavirin taken for 6 months to 1 year in order to attain a sustained viral response, namely undetectable virus 6 months after the completion of treatment8. Though expensive, the current HCV therapy is thought to be cost-effective, based on projections that sustained viral response may eliminate utilization of healthcare resources and associated healthcare costs associated with HCV-related liver disease9–11.

While previous studies have evaluated these impacts of HCV, they suffer significant limitations in their inability to separate out the independent contribution of HCV infection from many socio-demographic characteristics and comorbid disease states that are prevalent among patients with HCV. For example, HCV patients with psychiatric illnesses12–14, fatigue15, and depression symptoms16,17 have been shown to have worse health-related quality of life (HRQoL) than those without these comorbidities. Similarly, healthcare resource utilization in HCV patients is greater when HCV coexists with alcoholic liver disease.

This analysis attempts to overcome limitations of previous studies through the application of a propensity scoring model, which allows for the elimination of selection biases and explicit consideration of socio-demographic and comorbid disease characteristics that may impact HCV disease outcomes. As HCV is the most common chronic blood-borne infection in the US, this study aims to address the following research question: is the presence of diagnosed HCV associated with poorer HRQoL and greater healthcare resource use? The study findings will allow quantification of two main areas of disease burden for HCV infection: (1) economic: resource use and burden of illness; and (2) humanistic: HRQoL impact.

Methods

Sample: National Health and Wellness Survey

The study sample and data were taken from the 2009 wave (N = 75,000) of the US National Health and Wellness Survey (NHWS) (Kantar Health, New York, NY, USA), an annual, cross-sectional study of subjects aged 18 years or older. The primary objective of the NHWS is to provide a comprehensive database of epidemiological and treatment information, healthcare attitudes, behaviors, demographic and disease characteristics, and health-related outcomes. The demographic composition of the 2009 US NHWS sample is comparable to that of the US adult population as assessed by the March 2008 Current Population Survey of the US Census Bureau (see Appendix Table 1). Additional comparisons with NHWS and NHIS have been made elsewhere18.

A self-administered, Internet-based questionnaire was completed by a sample population identified through a web-based consumer panel. The consumer panel recruits its panel members through opt-in emails, co-registration with panel partners, e-newsletter campaigns, banner placements, and both internal and external affiliate networks. All panelists must explicitly agree to be a panel member, register with the panel through a unique email address, and complete an in-depth demographic registration profile. A stratified random sample procedure (using quotas based on gender, age, and race/ethnicity) was implemented to be reflective of the demographic composition of the general population. All subjects provided informed consent and the study was approved by Essex Institutional Review Board (Lebanon, NJ, USA).

Data elements

HRQoL (SF-12 and SF-6D)

The Medical Outcomes Study 12-Item Short Form Survey Instrument (SF-12) is a multipurpose, generic HRQoL instrument comprising of 12 questions19. The instrument is designed to report on eight health concepts (physical functioning, role physical, bodily pain, general health, vitality, social functioning, role emotional, and mental health). The SF-12 questions were selected from the SF-36 health survey20,21. In order to construct the shortest possible form that would provide comparable information as provided by the physical and mental component summary scores (PCS and MCS, respectively) and the eight health concepts of the SF-36, 12 out of the original 36 items in the SF-36 were selected. Details of how the links are established and the scoring algorithms are given in elsewhere19. For the purpose of the present analysis, the PCS and MCS summary scores are utilized as normed scores. This is achieved by transforming the raw scores for the items to a mean of 50 and a standard deviation of 10 for the US population. Normed scores can be calculated for both the eight SF-12 scales as well as for the PCS and MCS summary scores.

In addition to generating profile and summary PCS and MCS scores, the SF-12 can also be used to generate health state utilities. This is achieved through application of the SF-6D which takes 6 items from the SF-12. The SF-6D is a preference-based single index measure for health using general population values22,23. The 18,000 health states that the SF-6D is able to describe are correlated with preference weights obtained from a sample of the UK general population using the recognized standard gamble valuation technique. The SF-6D index has interval scoring properties and yields summary scores on a theoretical 0–1 scale (with an empirical floor of 0.203). The preference weights have recently been revised24.

Healthcare resource utilization

The 2009 NHWS also asks respondents about their use of healthcare resources. Resource utilization is considered in terms of the number of visits in the last 6 months to healthcare providers, the emergency room (ER), and the hospital for the patient's own medical condition. Healthcare providers include general practitioner/family practitioners, internists and dentists as well as more specialized physicians.

Predictor variables

A wide array of predictor variables was evaluated that have previously been shown to have potential impact on both HRQoL and healthcare resource utilization, which were grouped into socio-demographic variables, health risk behaviors, and morbidity/comorbidity status.

The socio-demographic variables included gender, race/ethnicity (white, black, Hispanic, Asian, or other), marital status (married/living with partner vs. all else), educational attainment (college degree vs. all else), employment status (currently employed vs. not employed) and annual household income (<$25 K, $25 K to <$50 K, $50 K to <$75 K, $75 K or more, decline to answer). Health risk behaviors included tobacco smoking, alcohol consumption, body mass index (BMI) and physical exercise. For the assessment of morbid/comorbidity status, both the number of comorbid conditions and, separately, the self-reported presence of anxiety and depression were assessed.

Statistical analysis

In order to identify members of a matched-control group to compare with the hepatitis C group, a propensity scoring methodology was employed. Specifically, age, gender, race/ethnicity, education, income, health insurance, employment status, presence of anxiety, presence of depression, number of comorbidities (excluding anxiety and depression), smoking, exercise, alcohol use, and BMI were included in a logistic regression to predict group assignment (HCV group vs. unmatched control group) Each HCV patient was matched with a control patient whose propensity score was nearest using a SAS macro (greedy matching algorithm). The greedy matching algorithm is one of the most widely used algorithms in propensity score matching analysis and allows for each case to be matched with the most suitable control available at that point in the matching process25,26. This is done by performing up to seven passes to find one matched controls for each case. First, the algorithm searches for a control with a propensity score within 0.0000001 of a case's propensity score value. If none is found, then the algorithm searches for a control within 0.000001 and continues searching for a suitable control with decreasingly-restrictive criteria (0.00001, 0.0001, 0.001, and 0.01) until a control is found25,26.

Differences between the HCV group and the matched control group were analyzed using chi-square tests for categorical variables and t-tests for continuous, normally-distributed variables. Generalized linear models, specifying a negative binomial distribution and a log-link function, were used to analyze group differences on non-normal outcomes (i.e., resource use variables). All analyses were conducted using SAS v9.1.

Results

In the NHWS study, 501,239 subjects were contacted, out of whom 92,759 responded (18.5% response rate). Of those who responded, 75,000 patients gave informed consent, met inclusion criteria, and completed the survey. Patients diagnosed with hepatitis B, human immunodeficiency virus, or acquired immune deficiency syndrome (n = 966) were excluded from the current study, leaving a total sample size of n = 74,034. As the base rates of these conditions are low among the general population yet relatively high among the HCV population, patients diagnosed with these conditions were excluded to ensure any observed differences between HCV diagnosed patients and controls were not due to HCV-related comorbidities.

A logistic regression was used to predict group assignment (HCV vs. control) from demographic and health characteristics (see Table 1). The overall model was significant (χ²(24) = 794.22, p < 0.0001; c-statistic = 0.79) and had a better fit than the null model (AIC = 7127.27 vs. 7874.49). Propensity score values for each respondent were saved from the logistic regression model (see Table 2). The propensity score values for the HCV group (mean = 0.0265, SD = 0.0289, range = 0.0008–0.2600) were significantly higher than the propensity score values for the control group (mean = 0.0092, SD = 0.0135, Range = 0.0002–0.4540; t(697) = −15.73, p < 0.0001).

Table 1.  Logistic regression statistics when predicting group membership of HCV or controls.

	b	SE b	χ^2	p-value	OR	95% LCL	95% UCL
Intercept	−5.4444	0.3115	305.5163	<0.0001
Age	0.0157	0.00311	25.5486	<0.0001	1.016	1.01	1.022
Female	−0.6654	0.0811	67.2602	<0.0001	0.514	0.439	0.603
White (reference)	–	–	–	–	–	–	–
Hispanic	0.384	0.143	7.2106	0.0072	1.468	1.109	1.943
Black	0.3536	0.1275	7.6874	0.0056	1.424	1.109	1.829
Asian	−0.8714	0.4161	4.385	0.0363	0.418	0.185	0.946
Other ethnicity	0.6199	0.1656	14.0134	0.0002	1.859	1.344	2.571
Married	−0.0492	0.0842	0.3406	0.5595	0.952	0.807	1.123
College educated	−0.1961	0.0977	4.0312	0.0447	0.822	0.679	0.995
Household income: < $25K	0.2618	0.1014	6.6688	0.0098	1.299	1.065	1.585
Household income: $25K to <$50K (reference)	–	–	–	–	–	–	–
Household income: $50K to <$75K	−0.0384	0.1157	0.1103	0.7398	0.962	0.767	1.207
Household income: $75K+	−0.1963	0.1293	2.3052	0.1289	0.822	0.638	1.059
Household income: Decline to answer	−0.2725	0.2222	1.5036	0.2201	0.761	0.493	1.177
Health insurance	0.1773	0.098	3.2735	0.0704	1.194	0.985	1.447
Currently employed	0.0168	0.088	0.0364	0.8487	1.017	0.856	1.208
Currently smoke	1.1161	0.0826	182.3839	<0.0001	3.053	2.596	3.59
Exercised in the last month	−0.1114	0.0804	1.9188	0.166	0.895	0.764	1.047
Alcohol use	−0.3508	0.0808	18.8252	<0.0001	0.704	0.601	0.825
BMI: underweight	−0.8536	0.4599	3.4456	0.0634	0.426	0.173	1.049
BMI: normal (reference)	–	–	–	–	–	–	–
BMI: overweight	0.0609	0.1041	0.3424	0.5585	1.063	0.867	1.303
BMI: obese	0.0671	0.1017	0.4357	0.5092	1.069	0.876	1.305
BMI: decline to answer	0.1482	0.6776	0.0478	0.8269	1.16	0.307	4.376
Comorbidity count	0.0692	0.00757	83.6865	<0.0001	1.072	1.056	1.088
Anxiety	0.2084	0.1111	3.5163	0.0608	1.232	0.991	1.531
Depression	0.4789	0.1063	20.2842	<0.0001	1.614	1.311	1.988

Table 2.  Propensity score values for the control and HCV groups both pre- and post-match.

Propensity score	Pre-propensity match				Post-propensity match
	HCV = 695		Control = 73339		HCV = 695		Control = 695
	n	%	n	%	n	%	n	%
0.0000 to <0.0125	288	41.44	59,096	80.58	288	41.44	288	41.44
0.0125 to <0.0250	164	23.60	9,255	12.62	164	23.60	164	23.60
0.0250 to <0.0375	84	12.09	2,634	3.59	84	12.09	84	12.09
0.0375 to <0.0500	56	8.06	1,020	1.39	56	8.06	57	8.20
0.0500 to <0.0625	30	4.32	510	0.70	30	4.32	29	4.17
0.0625 to <0.0750	24	3.45	292	0.40	24	3.45	24	3.45
0.0750 to <0.0875	23	3.31	169	0.23	23	3.31	23	3.31
0.0875 to <0.1000	9	1.29	112	0.15	9	1.29	9	1.29
0.1000 to <0.1125	4	0.58	82	0.11	4	0.58	4	0.58
0.1125 to <0.1250	3	0.43	47	0.06	3	0.43	3	0.43
0.1250 to <0.1375	3	0.43	31	0.04	3	0.43	3	0.43
0.1375 to <0.1500	1	0.14	24	0.03	1	0.14	1	0.14
0.1500 to <0.1625	0	0.00	17	0.02	0	0.00	0	0.00
0.1625 to <0.1750	3	0.43	13	0.02	3	0.43	3	0.43
0.1750 to <0.1875	1	0.14	8	0.01	1	0.14	1	0.14
0.1875 to <0.2000	0	0.00	8	0.01	0	0.00	0	0.00
0.2000 to <0.2125	1	0.14	6	0.01	1	0.14	1	0.14
0.2125 to <0.2250	0	0.00	1	0.00	0	0.00	0	0.00
0.2250 to <0.2375	0	0.00	3	0.00	0	0.00	0	0.00
0.2375 to <0.2500	0	0.00	3	0.00	0	0.00	0	0.00
0.2500 to <0.2625	1	0.14	0	0.00	1	0.14	0	0.00
0.2625 to <0.2750	0	0.00	1	0.00	0	0.00	1	0.14
0.2750 to <0.2875	0	0.00	1	0.00	0	0.00	0	0.00
0.2875 to <0.3000	0	0.00	1	0.00	0	0.00	0	0.00
0.3000 to <0.3125	0	0.00	2	0.00	0	0.00	0	0.00
0.3125 and greater	0	0.00	3	0.00	0	0.00	0	0.00

Table 3 compares respondent characteristics by the HCV status. Subjects with HCV diagnosis were older (mean = 50.4 year (SD 11.0) vs. 46.0 years (SD 16.9); p < 0.0001) and less likely to be female (41.9 vs. 52.2%, p < 0.0001) compared to those without. Fewer HCV subjects were married (54.5 vs. 59.4%, p = 0.017) and college‐educated (22.6 vs. 36.1%, p < 0.0001). The proportion with an annual income of less than $25,000 was lower in the HCV group (31.8 vs. 18.9%, p < 0.0001) relative to controls. Overall, HCV subjects had less healthy characteristics. For example, the prevalence of tobacco smoking, anxiety disorder and depression was more than two fold higher in the HCV group compared to the controls.

Table 3.  Respondent characteristics.

	HCV group n = 695				Unmatched control group n = 73,339				p-value
	Unweighted		Weighted		Unweighted		Weighted
Variable	n	n	%	SE	n	n	%	SE
Female	286	865,034	41.9	±1.98	37,964	114,086,430	52.2	±0.21	<0.00001
White	510	1,390,781	67.4	±2.03	54,412	150,917,684	69.0	±0.21	0.449
Black	76	223,617	10.8	±1.20	7,494	24,668,483	11.3	±0.14	0.7198
Hispanic	60	306,387	14.9	±1.84	5,947	28,768,182	13.2	±0.16	0.3604
Asian	6	16,465	0.8	±0.33	3,430	8,448,284	3.9	±0.08	<0.00001
Other race	43	125,310	6.1	±0.91	2,056	5,962,965	2.7	±0.08	0.0003
Married/living with partner	383	1,124,619	54.5	±2.02	44,978	129,990,087	59.4	±0.20	0.0169
College-educated	166	466,643	22.6	±1.65	26,887	78,897,185	36.1	±0.20	<0.00001
Employed	293	882,674	42.8	±1.99	39,497	122562298	56.0	±0.21	<0.00001
Have health insurance	536	1,582,740	76.7	±1.67	60,466	176,869,058	80.8	±0.16	0.0145
Income <$25k	225	655,725	31.8	±1.84	13,299	41,247,829	18.9	±0.17	<0.00001
Income $25k to <$50k	224	672,893	32.6	±1.89	21,952	65,443,921	29.9	±0.19	0.1543
Income $50k to <$75k	123	373,812	18.1	±1.67	15,749	46,376,502	21.2	±0.16	0.0665
Income $75k+	100	294,901	14.3	±1.39	17,884	52,614,301	24.1	±0.17	<0.00001
Decline to Answer	23	65,228	3.2	±0.67	4,455	13,083,045	6.0	±0.10	<0.00001
Currently smoke	358	1,114,186	54.0	±2.00	16,286	49,236,764	22.5	±0.17	<0.00001
Drink alcohol	379	1,141,257	55.3	±2.00	47,698	143,079,030	65.4	±0.20	<0.00001
Currently exercise	357	1,061,104	51.4	±2.02	46,547	140,889,633	64.4	±0.20	<0.00001
Comorbid anxiety	251	756,353	36.7	±1.92	11,451	35,110,320	16.0	±0.15	<0.00001
Comorbid depression	294	885,238	42.9	±1.99	12,907	38,960,073	17.8	±0.15	<0.00001
	Unweighted		Weighted		Unweighted		Weighted
	Mean	Mean	SD	SE	Mean	Mean	SD	SE
Age	51.46	50.35	11.02	0.42	47.89	45.96	16.93	0.06	<0.00001
Body mass index (BMI)	29.71	29.78	7.45	0.28	28.78	28.62	7.04	0.03	<0.00001
Number of comorbidities	8.14	8.25	6.29	0.24	4.34	4.18	4.49	0.02	<0.00001

After the propensity score match process (which identified a single control for each case), the propensity scores of the HCV group (mean = 0.0265, SD = 0.0289, range = 0.0008–0.2600) were not significantly different than the matched controls (mean = 0.0265, SD = 0.0289, range = 0.0008–0.2629; t (1388) = −0.00, p = 0.9967; see Table 2). As expected, post-propensity matching there were no differences between the HCV group and the matched control group on any of the variables.

Table 4 summarizes the HRQoL data. When compared to propensity score matched controls, respondents with HCV reported significantly worse scores in six of the eight health dimensions including bodily pain, general health, physical functioning, physical role limitations, social functioning, and vitality. The other two domains, namely mental health and emotional role limitations also showed decrement in HRQoL among HCV patients, although the difference did not reach statistical significance. The difference between the HCV and matched controls was larger for the PCS score than the MCS, the latter not being significant. Health utilities estimated by SF-6D preference scores were 0.63 for the HCV group and 0.66 for the controls. The data in the unmatched controls showed uniformly and significantly more favorable assessment of HRQoL.

Table 4.  Impact of HCV on health related quality of life (weighted).

	HCV group n = 695		Unmatched control group n = 73339			Matched control group n = 695
	Mean	SE	Mean	SE	p-value*	Mean	SE	p-value†
Mental component summary (MCS)	42.70	0.46	47.36	0.48	<0.00001	43.20	0.48	0.45
Physical component summary (PCS)	39.57	0.48	47.94	0.49	<0.00001	42.66	0.49	<0.00001
Bodily pain	53.03	1.19	74.36	0.10	<0.00001	60.96	1.18	<0.00001
General health	48.15	1.06	68.42	0.09	<0.00001	55.82	1.05	<0.00001
Mental health	55.94	0.94	66.12	0.08	<0.00001	57.18	0.95	0.35
Physical functioning	56.46	1.37	75.98	0.12	<0.00001	61.57	1.42	0.0097
Role emotional	65.45	1.11	78.39	0.10	<0.00001	66.60	1.17	0.48
Role physical	51.64	1.24	72.44	0.11	<0.00001	56.96	1.27	0.0028
Social functioning	57.89	1.16	76.06	0.10	<0.00001	62.73	1.24	0.0044
Vitality	38.56	0.97	51.57	0.09	<0.00001	43.46	1.02	0.0005
Health utility	0.63	0.01	0.73	0.00	<0.00001	0.66	0.01	0.0044

*Indicates the significance comparing the HCV group with the unmatched control group.

†Indicates the significance comparing the HCV group with the matched control group.

Table 5 summarizes the distribution of visits reported for the HCV and control populations. Compared to the unmatched controls, HCV subjects had more than twice the number of emergency room and physician visits and hospitalizations. When the matched controls were considered for comparison, the differences were reduced but both the number of emergency room visits and hospitalizations were significantly greater among HCV subjects.

Table 5.  Impact of hepatitis C on healthcare resource utilization (weighted).

	HCV group n = 695		Unmatched control group n = 73,339			Matched control group n = 695
	Mean	SE	Mean	SE	p-value*	Mean	SE	p-value†
Emergency room visits	0.59	0.06	0.22	0.00	<0.00001	0.44	0.05	0.042
Hospitalizations	0.30	0.04	0.12	0.00	<0.00001	0.20	0.04	0.075
Traditional provider visits	7.71	0.39	3.91	0.02	<0.00001	5.93	0.28	0.0002

*Indicates the significance comparing the HCV group with the unmatched control group.

†Indicates the significance comparing the HCV group with the matched control group.

Discussion

The primary purpose of the present study was to investigate whether the presence of diagnosed HCV negatively impacts HRQoL and increases resource use in the US population. Although previous studies have documented a lower level of health-related quality of life among HCV patients compared to controls, many of them were based on tertiary care referral patients that may not be representative of HCV patients at large in the population. Other studies failed to appreciate the potential confounding effects of socio-demographic, health-risk behavior and comorbidity profiles in HCV patients, which may exaggerate the difference between subjects with HCV and those without. Based on our logistic regression results, there were many differences between HCV patients and controls; all of which were eliminated in our propensity score matching process. As such, our data clearly demonstrated that (1) without a proper matching, the comparison between HCV and non-HCV subjects does not provide the accurate assessment of the impact on HRQoL and healthcare resource utilization attributable to HCV and (2) when appropriate control subjects were used, HCV still has a significant independent impact on the quality of life and healthcare utilization of the individual. In particular, the present analysis confirms the independent effect of HCV over and above comorbidities, including anxiety and depression. For example, although anxiety and depression are common and lead to HRQoL discrepancies between HCV patients and non-patients, the data point clearly to a statistically and clinically meaningful independent effect of HCV.

There are a number of strengths to the NHWS data. First, the sample is representative of the US population maximizing the generalizability of the results to HCV patients in the US. Second, the database was large enough to include a sufficient number of subjects with HCV and thus to give adequate power to the study. Third, it entails a large number of variables that may be expected to impact both HRQoL and healthcare resource utilization for computation of a propensity score for selection of proper controls for the study. The propensity scoring approach that was used for the analysis generates effect estimates for HCV based on a large number of measured contextual variables. While it would be possible to utilize alternative techniques, such as instrumental variables, to assess and account for the potential impact of selection bias in single equation regression models, such models are susceptible to generating biased and inconsistent parameter estimates of the effect of HCV. In contrast, the propensity scoring approach is most widely used technique in the clinical sciences to generate average effect of intervention from an observational data set. In this study, the method of analyzing the effect of HCV on health outcomes in this cross sectional data set was adapted.

The results presented here should be seen against the background of a growing literature on the impact of HCV. Over the past 15 years there have been a number of studies that have attempted to assess the impact of HCV, together with hepatitis B and cirrhosis of the liver, on HRQoL and, to a lesser extent on healthcare resource utilization. While these studies point to the negative impact of HCV, the extent to which HCV has an independent effect reflect a number of study characteristics, such as the choice of database or HRQoL instrument and patient characteristics such as socio-demographic profile, the presence of comorbidities, and possible treatment interventions. In consequence, it is difficult to come to a firm conclusion as to the actual quantitative impact of HCV and its relation to the potential impact of other factors on HRQoL and healthcare resource utilization. Bonkovsky and Woolley (1999) found that HCV patients scored significantly lower on SF-36 subscales when compared to population norms3. HCV patients reported significantly lower levels of quality of life relative to healthy controls, especially among the domains of depression, fatigue, vitality, and social and cognitive function2,4,5. The results of the present study now further extend the observation by utilizing the matched control design. Having explicitly controlled for anxiety and depression, the present study clearly showed that there still is a substantive impact of HCV on bodily pain, general health, physical functioning, social functioning and vitality.

Health utilities are an important component in health economic evaluations. In this study, SF-6D utility score for HCV patients was 0.63, which may be interpreted to estimate that HCV subjects feel that their current life is only worth 63% of one with perfect health. While the difference between the HCV and matched control subjects (0.63 vs. 0.66, p < 0.01) appears to be relatively small, 0.03 difference is considered to be clinically meaningful when measuring within-patient differences27 and must be considered in future cost-effectiveness analyses. These results are also consistent with those reported by Dan et al. (2008) where, controlling for the impact of gender, age and cirrhosis, the impact of chronic liver disease on SF-6D utility scores was substantially lower for HCV than for hepatitis B patients28.

The present analysis also confirms the impact of diagnosed HCV on healthcare resource utilization – in terms of both provider and emergency room visits. Although the present study focused resource use visits, the results have clear economic implications (i.e., HCV patients would invariably have higher direct costs). Estimated direct healthcare costs related to HCV amounted to $2,070 per patient in 19979. Later analyses found a median of $2,470 per patient in the time period of 1997–1999 for direct healthcare costs related to HCV11. However, many of the studies employed a ‘top-down’ approach to estimating the resource use impact of HCV. The current study relied on a patient perspective (a ‘bottom-up’ approach) and provided current data, as opposed to projections from older sources. It is also worth noting that many of the alternative explanations citied for the HRQoL findings (e.g., psychiatric and physical comorbidities) were controlled for in the resource use models, so the differences could be assumed to be the isolated effect of HCV. Previous studies have not, however, taken a patient perspective in evaluating the impact of HCV. Nor have previous studies, utilizing recall estimates, attempted to differentiate the independent impact of HCV on provider visits (almost 40% more), emergency room visits and hospitalizations. The present study points to the importance of such a distinction given the lack of any discernible impact of HCV on the number of hospitalizations.

The current study provides helpful insights into a variety of factors that may influence HRQoL in general and in HCV patients, further justifying our propensity score matching method. First, the relationship between age and HRQoL and healthcare resource utilization is well established. National population surveys such as the Behavioral Risk Factor Surveillance System in the US have shown that, on a range of measures, HRQoL declines with increasing age while healthcare resource utilization increases. Second, the presence of morbid/comorbidity factors are expected to impact both HRQoL and healthcare resource utilization. For example, the impact of BMI on HRQoL and healthcare resource utilization is also well known. A recent paper by Søltoft et al. (2009), utilizing data from the 2003 Health Survey of England, found a significant association between BMI and HRQoL29. The study found that after controlling, among other variables, for gender, age and obesity related comorbidities, HRQoL was at a maximum with a BMI of 26.0 in men and 24.5 in women. There was also a negative association for both underweight and overweight individuals. Third, with regard to health-risk behaviors, following an exercise regimen has been shown to positively impact HRQoL30.

Alcohol consumption is not only an important co-factor in HCV disease progression but also an important morbidity on its own and a determinant of HRQoL. Assessing the impact of alcohol consumption on HRQoL is dependent upon the measures of alcohol consumption used. Evidence to date would suggest a non-linear relationship31. Moderate drinking is associated with similar or higher HRQoL scores compared to non-drinkers. Substantial HRQoL deficits are associated with higher levels of daily alcohol consumption and binge drinking. The picture is further clouded once former drinkers are included in the assessment32. In NHWS, alcohol consumption was defined as current use versus non-use, which may explain the lower prevalence of current alcohol use in the HCV group. The lack of quantitative assessment of alcohol consumption makes it difficult to project a relationship with either HRQoL or resource utilization. The relationship between smoking and HRQoL is a bit more controversial. A short summary of the current data may be that smoking is expected to have a negative, but probably small, impact on HRQoL in patients with HCV.

Limitations

Due to the propensity scoring methodology employed here, none of the effects observed could be attributed to any demographic or health history variables included in the analyses. Of course, it is possible that there may be additional variables not included, which could explain the observed differences in health outcomes. This is an important limitation of the current study. However, the most likely factors (comorbidities, health behaviors, etc.) have been accounted for. It is also important to note that because of the survey methodology, it was not possible to verify HCV diagnosis. Nevertheless, many of the findings coincide with that of the literature, suggesting our HCV sample is similar to that of other, clinically-verified HCV samples. Further, if actual HCV patients were in the control group (because they were not aware of their disease status) then the current study results likely underestimate the true effect of HCV. Aside from verified diagnosis, the current study did not assess reasons for healthcare resource utilization but, given the propensity score methodology, the assumption was made that the additional resources used by the HCV group were due to the virus itself (since all other confounders were held constant). The survey also did not assess other clinical details of the patients (e.g., cirrhosis, hepatocellular carcinoma), which should be included in future studies.

A final limitation to consider is the Internet-survey methodology. It remains unclear if patients who did not complete the survey differ in meaningful ways from those who did. However, given the similarities between NHWS and other established sources (see Appendix Table 1), it would appear as if the overall NHWS is representative of the US population at large. It should also be noted that sampling weights were applied to all analyses in order to correct for any sampling bias inherent in the NHWS.

Conclusions

The presence of diagnosed HCV in the US adult population has been shown, utilizing a propensity scoring model, to have a significant and clinically relevant impact on key dimensions of HRQoL. At the same time, the presence of HCV, utilizing the same propensity scoring model, has been shown to have an impact on the utilization of traditional provider and ER visits within the healthcare system. There is no apparent impact on reported hospitalization experience. These results, in pointing the independent effect of HCV, confirm previous assessments of the impact of this disease – although, as noted, previous modeled claims suffer from potentially significant methodological limitations. These analyses accounted for an exhaustive array of demographic and health history characteristics (such as anxiety and depression) which are known to burden the HCV population and be associated with health outcomes.

Application of propensity scoring, however, while proving a methodologically robust and widely accepted approach to assessing disease and treatment effects from observational data, says nothing as to the relative impact of HCV vis à vis the contribution of associated socio-demographic characteristics, the presence of health-risk factors and other morbid or co-morbid conditions on HRQoL and healthcare resource utilization. Further analysis, utilizing for example an instrumental variable approach, could address these issues at the patient level. This assumes that it would be possible to develop an appropriate instrument with data from the NHWS to predict the presence of HCV. Such an approach would not only address the question of the quantitative importance of selection bias on treatment effect estimates in HCV, but would also provide the basis for an assessment of potential interactions between the presence of HCV and other confounding factors in the societal burden impact of this highly prevalent disease state across multiple domains.

Finally, the current study found the number of emergency room visits and hospitalizations to be significantly greater among HCV subjects when compared to matched controls. However, the reliance on self-reported resource use data may have introduced recall bias and costs were not included as part of NHWS. Future research is needed to quantify the total economic impact diagnosed HCV has on the US healthcare system.

Transparency

Declaration of funding

The National Health and Wellness Survey (NHWS) is conducted by Kantar Health. Bristol-Myers Squibb purchased access to the NHWS dataset and funded the analysis for this project.

Declaration of financial relationships

Both M.D. and S.W. are employees of Kantar Health while Y.Y. and G.L. are employees of Bristol-Myers Squibb. P.L. served as a consultant to Kantar Health.

Appendix

Appendix Table 1.  Comparison of the demographic profile of NHWS respondents and the US adult population.

Demographic profile	National Health and Wellness Survey (NHWS) 2009 %	US reference %*
Gender
Male	49	48
Female	51	52
Age
18–44	43	49
45–64	34	34
65+	22	17
Race/ethnicity
African-American	10	11
Hispanic	8	13
White	74	69
Other	7	6
Annual household income (US$)
<$25,000	20	21
$25,000–49,999	32	26
$50,000–74,999	23	20
$75,000+	26	33
Education
No college degree	63	73
College degree	37	27
Region
Northeast	18	18
Midwest	24	22
South	35	36
West	22	23
Morbidity prevalence†
Hypertension	29.4	25.0
Stroke	1.2	2.9
Emphysema	1.5	1.7
Asthma	8.6	12.6
Hay fever	8.6	8.0
Chronic bronchitis	4.6	4.4
Any cancer	8.3	7.9
Breast cancer	1.5	1.4
Cervical cancer	0.9	1.4
Prostate cancer	1.0	2.0
Diabetes	10.7	8.4
Ulcers	2.9	8.6
Kidney disease	0.8	1.7
Arthritis	19.9	22.8
Migraine/severe headache	11.9	13.4

*Current Population Survey (2008) of the US Census Bureau except the morbidity prevalence data.

†The US reference is according to the National Health Insurance Survey (Center for Disease Control and Prevention). NHWS data weighted by age, gender, and race/ethnic groups.