Introduction

Chronic hepatitis C virus (HCV) infection affects over 3 million Americans and is a leading cause of liver failure, cirrhosis, and liver cancer[1, 2]. Though primarily associated with liver disease, recent evidence suggests that HCV is a multi-faceted systemic condition that may be linked to many extra-hepatic disorders (EHDs) including arthritic-like pain, endocrine, metabolic, kidney, neuropsychiatric, and cardiovascular conditions [3–5]. Studies conducted during the interferon (IFN) treatment era found associations between HCV and neuropsychiatric, somatic and gastrointestinal (GI) symptom clusters, most commonly fatigue, sleep disturbance, irritability, depression, and pain[6–11]. Symptoms may be attributable to liver disease or to underlying inflammatory processes that mediate the relationship between HCV infection and EHDs [4, 12]. Beyond disease-related factors, psychosocial factors may also contribute to symptom burden. Psychiatric and substance use disorders, salient risk factors for contracting HCV, are highly prevalent among HCV patients and may be directly associated with poor health outcomes, irrespective of HCV infection or liver disease[9, 10, 13–15]. Indeed, neuropsychiatric symptoms are reported in up to 50% of HCV patients, independent of liver disease severity[16]. Additionally, the HCV population is further disadvantaged by social determinants of poor health (SDoH), including economic instability, low income, unemployment and lack of health insurance[17–19]. Finally, the psychological sequelae of harboring a transmittable disease, social stigma, and anxiety related to deteriorating liver health may also contribute to symptom burden and poor health outcomes[20–24]. Fortunately, the impact of these factors on health outcomes may be mitigated by public awareness that a short course of well tolerated direct acting antiviral (DAA) therapy has over 95% efficacy in achieving virologic cure.

Much of the data on HCV-associated symptoms, health-related quality of life (HRQOL), and treatment side effects, comes from the IFN treatment era [6–8, 25]. Qualitative studies elucidated patients’ experiences of symptoms and HRQOL[7, 26–28], but far fewer quantitative studies provided any in-depth analysis of HCV symptoms, especially among patients not on IFN therapy[6, 7]. During the IFN era, researchers identified 22 key patient-reported outcome (PRO) concepts from qualitative studies as important to the HCV population[7]. The majority of these PRO concepts received inadequate attention then and have received virtually no attention during the current DAA era. Several recent studies of HCV-infected patients treated with DAAs have investigated a few PRO concepts, such as HRQOL, fatigue, and work productivity[29, 30]. However, these data were derived from drug registration trials that enrolled highly selected patients with limited participation of under-represented racial minorities and subgroups with decompensated liver disease, EHDs, and medical, psychiatric and addiction comorbidities.

A comprehensive understanding of baseline symptom burden in patients with HCV is necessary to lay the groundwork for subsequent real-world investigations of potential changes in symptoms during DAA therapy and after virologic cure. We aimed to characterize patient-reported symptoms, medical conditions, and functional well-being in a large multi-center US cohort who initiated DAA therapy in clinical practices in 2016-2017. Our secondary aim was to evaluate sociodemographic/SDoH, liver-related, and other clinical features associated with these health outcomes.

Methods

Statistical analysis strategy

Descriptive statistics.

To characterize the study cohort, graphical and tabular descriptive statistical methods (means, standard deviations (SD), range) were used to visualize the data and describe the empirical distributions of the four PRO constructs: (1) Number of medical comorbidities; (2) Specific Symptom Clusters (PROMIS instruments); (3) Overall Symptom Burden (TMSAS); and (4) Functional Well-Being (HCV-PRO), as well as participant characteristics (sociodemographics, liver-related features, psychiatric and substance abuse history).

Bivariate associations.

To help characterize relationships between each patient characteristic and each PRO, unadjusted point and interval estimates of correlation coefficients were tabulated along with unadjusted estimates of subgroup PRO means/medians. Pearson correlation coefficients were used for continuous-vs-continuous pairs of variables; point-biserial correlation coefficients were used for binary (e.g. gender, cirrhosis)-vs-continuous pairs; Spearman correlation coefficients were used for ordinal (e.g., cirrhosis status combined with MELD score)-vs-continuous pairs; and for continuous PRO measures and multi-level nominal categorical variables (e.g., race, employment), we used the square root of the unadjusted R² for the general linear model of the PRO measure conditional on the categorical variable as a predictor. The square root of R² is a non-directional absolute value (r).

Multivariable models of association.

In the inferential investigation of patient characteristics that may be predictive of PROs, we implemented a cross-validation strategy. Participants were assigned by randomization to two groups: sample 1 or sample 2. Sample 1 (n = 700) was used for exploratory model-building analyses to generate a set of candidate predictor variables that might be associated with each outcome. Sample 2 (n = 900) was used to test and potentially validate the candidate models. Essentially, two identical studies were performed in which the second was used to validate predictor models generated by the first.

Based on documented and informal information from the data collection process, missing values of the outcome variables and the candidate predictor variables are presumed missing at random or missing completely at random. Complete-case analyses (omitting patients with missing data) suffers from selection bias and loss of precision. Therefore, missing values for the predictor variables were addressed via multiple imputation. Patients having a missing value for an outcome variable were omitted from that regression analysis. A multivariate multiple imputation algorithm (SAS procedure MI) was used to generate 40 completed copies of the dataset for the multivariable analyses. Each statistical regression model of interest was fitted to all 40 datasets. The 40 sets of results were combined (SAS procedure MIANALYZE) to produce the final results shown for each multivariable regression model.

The candidate predictor variables for model-building based in Sample 1 are found in Table 1 and include: (1) sociodemographics; (2) liver-related clinical and laboratory markers; and (3) mental health and substance abuse variables. Some of the laboratory markers were transformed to log₁₀ scale. In using Sample 1 to select candidate multivariable models for each PRO, we first examined models accounting for sociodemographics, then we looked at the additional contribution of liver lab and clinical variables, and finally psychiatric and substance abuse variables were included. Model-building with Sample 1 relied on stepwise variable selection algorithms or use LASSO (least absolute shrinkage and selection operator) variable-selection algorithms. Having completed all exploratory analyses and model development with Sample 1 (n = 700), we then used the data from Sample 2 (n = 900) for validation. Candidate predictor variables in models based on Sample 2 were considered “validated” if their regression coefficients were statistically significant at level α = 0.01.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Patients characteristics (n = 1600).

https://doi.org/10.1371/journal.pone.0196908.t001

Model-based analyses for the patient-reported ‘Number of Medical Comorbidities’ relied on generalized log-linear models representing a mean count of comorbid conditions as a function of patient characteristics. Similarly, multivariable analyses for TMSAS and HCV-PRO measures relied on generalized linear models. The PROMIS Fatigue and PROMIS Sleep Disturbance T-scores also were studied using generalized linear models because they exhibited symmetric discrete distributions.

In contrast, the other eight symptom-specific PROMIS T-scores exhibited semi-continuous empirical distributions; that is, the scores follow a discrete distribution with a clumping of values at the floor of each scale (patients who reported no symptom). Therefore, the analysis of each of these 8 PROMIS measures relied on a generalized linear model for zero-inflated Poisson distributions. To represent both the probability of having the symptom and the conditional mean score as functions of patient characteristics, the zero-inflated Poisson model specifies two simultaneous regression equations for the modified T-score: one for the distribution of non-zero values (patients who have the symptom) and one for the clump of values at zero (for patients without the symptom). The dependent variable was computed as (T-score − K) with K being the observed floor of the scale. The floor values (K) were: Depression score > 42; Anger score > 35; Anxiety score > 45; Cognitive Concerns > 25; Pain Interference > 45; Belly Pain > 30; Diarrhea > 32; Nausea/Vomiting > 37 (these floor values were also used for other analyses of dichotomized T-scores). In these analyses, we focused on identifying patient characteristics associated with the non-zero values (having the symptom) to identify patients at risk for high symptom burden.

For all the PROs, sensitivity analyses included variations on the type of model fitted, variations on the engine used for exploratory hypothesis generation (e.g. stepwise, LASSO, least-angle regression, model averaging) and variations on the assumptions and criteria used (e.g. link function, α-levels for entry and exit). For the PROMIS T-scores, sensitivity analyses included logistic regression for the dichotomized T-scores, cumulative logit model analysis for categorized T-scores (low, medium, high), and zero-inflated negative binomial model analysis. All statistical estimates were computed along with corresponding 95% confidence intervals. Statistical computations were performed using SAS System software version 9.4 (SAS Institute, Cary, NC). PROMIS T-scores were computed using R software, version 3.1.2 (2014 The R Foundation for Statistical Computing), and RStudio software, version 1.0.136 (RStudio Inc.).

Results

The study flowchart, which includes the number of patients screened, consented, and enrolled in PROP UP, is displayed in Fig 1. Of the nearly 2,400 patients screened, 87% consented to participate in the study. Of the 2082 consented patients, the main reasons patients were not enrolled included incomplete baseline surveys at the time that treatment was initiated (6%); insurance payer denials of DAA therapy (5%); patient lack of follow through with clinical requirements to start DAA therapy (e.g., completion of paperwork, urine toxicology screens) (3%); and verbal withdrawal from the study after consent (3%).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Study flowchart.

https://doi.org/10.1371/journal.pone.0196908.g001

To explore the potential for selection bias that could affect generalizability of the study results, we compared the 1600 enrolled patients with 757 patients who were screened or consented, but not enrolled, on age, sex, and race. Compared to the 1600 patients enrolled, the 757 screen/enrollment failures were similar in age (55 years versus 57 years), sex (61% male versus 55%), and race (59% White versus 61%; 29% Black versus 33%).

Number of medical comorbidities and association with patient characteristics

In addition to liver-related problems, patients reported the presence of current and past comorbid conditions, 97% of these conditions was endorsed by at least one participant. The 10 most frequently reported comorbidities are listed in Table 2. Patients endorsed a mean of four comorbidities (range: 0–15) (S2 Table). Less than 10% (n = 154) reported no medical comorbidities, thus the majority had at least one medical condition other than chronic HCV. Notably, many of these comorbid conditions could represent EHDs.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Top 10 patient reported medical comorbidities.

https://doi.org/10.1371/journal.pone.0196908.t002

As shown in Table 3, a greater number of medical conditions correlated with sociodemographics (.20 ≤ r ≤ .40; being older, low income, public health insurance, disability), liver-related markers (.17 ≤ r ≤ .18; MELD ≥ 12, low albumin, low hemoglobin, high creatinine), and mental health markers (15 ≤ r ≤ .24; psychiatric medication use, diagnosis, treatment, inpatient hospitalization). Correlations with alcohol and substance abuse were lower (0 ≤ r ≤ .09).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Patient characteristic subgroup means and correlations with PRO measures.

https://doi.org/10.1371/journal.pone.0196908.t003

Table 4 shows the results of the confirmatory multivariable analyses (n = 900) predicting Number of Medical Comorbidities and other PROs. The estimates are the predicted percent (%) increase in the mean score attributable to a specific patient characteristic. For example, a higher Number of Medical Comorbidities was associated with being older (2.7% higher for each additional year) and being disabled (53% higher).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. Validated predictors of PROs.

https://doi.org/10.1371/journal.pone.0196908.t004

Specific symptom clusters and association with patient characteristics

Fig 2 displays histograms for the T-scores for each of the PROMIS symptoms. Eight PROMIS measures showed a bimodal distribution with some patients reporting no symptoms while others reported mild to severe symptoms. T-scores over 55 are 1/2 standard deviation worse than the general US population and considered a clinically meaningful difference in other medical populations[34, 46–48].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Histograms of PROMIS symptom T-scores.

*The vertical line in each histogram shows the proportion of patients reporting no symptoms or responses at the minimum score.

https://doi.org/10.1371/journal.pone.0196908.g002

Neuropsychiatric symptom cluster.

Approximately 60-80% of patients endorsed symptoms in the neuropsychiatric symptom cluster (Fig 2). Bivariate correlations and unadjusted mean group differences are summarized in Table 3. Of the sociodemographic variables, the largest correlations with this symptom cluster were with employment (.19 ≤ r ≤ .28), health insurance (.13 ≤ r ≤ .19), and income (.09 ≤ r ≤ .18), such that disability, unemployment, low income, and public health insurance were associated with worse symptoms. Correlations between the neuropsychiatric symptom cluster and liver-related markers were quite small (0 ≤ r ≤ .07).

In multivariable models, severity of the neuropsychiatric symptom cluster was frequently associated with being disabled, unemployed, and use of psychiatric medications (Table 5). Patients who were disabled, unemployed, white, single and with past or current psychiatric medication use had 6% to 18% higher depression scores. Anger severity decreased with age and was 4% to 8% higher in patients who were disabled, unemployed, using psychiatric medications and abusing prescription drugs. Anxiety severity decreased with age and was 6% to 11% higher in patients who were disabled, of other race, uninsured, taking psychiatric medications, or had a history of psychiatric diagnosis. Severity in cognitive concerns was 4% to 9% higher in patients who were disabled, unemployed, on psychiatric medications, and history of psychiatric hospitalization.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 5. Validated predictors of T-score symptom severity.

https://doi.org/10.1371/journal.pone.0196908.t005

Somatic symptom cluster.

Almost all patients endorsed some level of fatigue and sleep disturbance, evidenced by the histograms in Fig 2 and 65% reported pain interference. Bivariate correlations and unadjusted mean differences between the somatic symptom cluster and patient characteristics are summarized in Table 3. In general, the largest correlations with the somatic symptom cluster were the sociodemographic and psychiatric variables. Patients who had lower income, were unemployed or disabled, or on public health insurance had worse somatic symptoms (.13 ≤ r ≤ .42) Patients with psychiatric markers were also observed to experience worse pain, sleep, and fatigue (.11 ≤ r ≤ .29). By contrast, correlations between the somatic symptom cluster and liver-related markers were small (.01 ≤ r ≤ .11).

Multivariable models demonstrated that fatigue severity was worse among patients who were female (6% higher), White or other race (9%-10% higher), disabled (9% higher) and those taking psychiatric medications (7% higher) (Table 4). Sleep disturbance severity was higher in patients who were unemployed (10% higher), disabled (11% higher), using psychiatric medications (7%-8% higher) (Table 4). Patients with lower income (5% higher), disabled (13% higher), unemployed (8% higher), with psychiatric medication use (5% higher) reported more severe pain interference (Table 5). No liver-related markers were identified as validated predictors of the somatic symptom cluster.

Gastrointestinal symptom cluster.

Approximately half of the cohort reported symptoms in the GI cluster, including abdominal pain (50%), nausea/vomiting (50%), and diarrhea (40%) (Fig 2). Bivariate correlations between the GI symptom cluster and patient characteristics are summarized in Table 3. Correlations were smaller with the GI cluster than with other symptom clusters. In general, the sociodemographic variables had the largest correlations with the GI cluster (.07 ≤ r ≤ .27), followed by correlations with the psychiatric markers (.10 ≤ r ≤ .20), and liver-related markers (0 ≤ r ≤ .11).

In multivariable models, nausea/vomiting severity was worse in patients who were disabled, unemployed or abusing prescription medications (14% to 23% higher) (Table 5). Abdominal pain severity was worse in patients who were disabled, unemployed, and with psychiatric diagnosis and psychiatric medication use (5% to 18% higher). Diarrhea severity was worse in patients who were disabled, taking psychiatric medications, and with a history of psychiatric diagnosis and hospitalization (7% -19% higher).

Overall symptom burden and associations with patient characteristics

On the MSAS, individual symptoms reported by at least 25% of participants are listed in Table 6 (S3 Table). The median TMSAS score was 0.4 (mean = 0.6, SD = 0.5, range: 0–3.1). The most common symptoms were lack of energy, pain, and difficulty sleeping. These symptoms were also the most severe and frequent.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 6. MSAS symptoms reported by at least 25% of patients.

https://doi.org/10.1371/journal.pone.0196908.t006

As shown in Table 3, higher overall symptom burden (higher TMSAS scores) was correlated with sociodemographics (.15 ≤ r ≤ .36; being female; lower income; disabled; uninsured; having public health insurance), lab biomarkers (.12 ≤ r ≤ .16; lower albumin, lower hemoglobin) and psychiatric markers (.23 ≤ r ≤ .35). Correlations with alcohol and substance abuse were lower (0 ≤ r ≤ .12). In multivariable regressions, symptom burden was higher in patients who were female (13% higher), disabled (41% higher), unemployed (29% higher), and taking psychiatric medications (33% higher) (Table 4).

Discussion

With recent advances in the treatment of HCV using highly potent DAAs and resultant changes to demographics of patients initiating DAA treatment, a contemporary and comprehensive assessment of the symptom burden of HCV-infected individuals in a real world setting, outside of carefully selected clinical trials populations, was needed. Towards this end, we conducted an in-depth analysis of patient-reported symptom burden including HCV-associated specific symptoms, medical comorbidities and functional well-being in a large multicenter cohort initiating DAA therapy. We also took the opportunity to thoroughly describe the cohort with regard to several sociodemographics, liver-related features, and mental health and drug and alcohol use parameters of the current treatment population. Through a rigorous analytical approach, we identified and validated key patient characteristics associated with each PRO. Our analysis has three key findings. First, many patients with chronic HCV, although not all, had a large number of medical comorbidities and high symptom burden, with the most common being fatigue, sleep disturbance, pain interference and neuropsychiatric symptoms. Second, most PROs were strongly associated with just a handful of patient characteristics; namely, disability, unemployment and current use of psychiatric medications. Other predictor variables for each PRO were identified, but with lower frequency and strength of association. Third, laboratory biomarkers and clinical markers of liver disease severity were not strong predictors of most PROs. These findings will lay the groundwork for subsequent longitudinal investigations of symptoms and comorbid conditions that may change over the course of DAA therapy and after viral eradication.

This study provides extensive sociodemographic information regarding the chronic HCV population currently being treated in the US. The study cohort was 61% White and 39% non-White (33% identified as Black/African-American, 6% identified as Other Race), in contrast to industry-sponsored clinical trials that under-represent minority populations (i.e., majority are >80% White (range: 66%-97%)[49–51]. The vast majority of patients initiating treatment had low educational attainment (not exceeding high school) and low household income (less or equal to $40,000 per year), well below the average income cited in the 2016 Census survey[52]. Over 50% were recipients of disability benefits, applying for disability benefits, or unemployed and looking for work. The majority was receiving public health insurance or was uninsured. Employment status, including disability and unemployment, as well as low income and lack of insurance were common patient features associated with high symptom burden and comorbidities, consistent with the larger literature on social and economic determinants of poor health in other medical populations). Recognizing the impact of SDoH on the health and treatment outcomes of the HCV population has both clinical and health policy implications. While the emphasis has been on viral eradication to improve liver-related outcomes, our results highlight the need to acknowledge the high burden of concomitant chronic health conditions among the chronic HCV population. In fact, HCV therapy initiation presents a window of opportunity for patients to re-engage with the healthcare system and address other health conditions. Improving the overall health of the HCV population will require improvements and innovation at multiple levels of healthcare including enhanced coordination of care between hepatologists and other subspecialists, multidisciplinary teams, co-location of mental health and addiction specialists, patient navigation, and innovative telehealth models [21, 53–63].

Nearly half of the cohort had cirrhosis and 12% had evidence of advanced cirrhosis (MELD ≥ 12). We suspect that state Medicaid restrictions during the enrollment period partially influenced variability of liver disease severity in this cohort by limiting access to treatment to patients with higher stages of fibrosis and denying treatment to those with minimal fibrosis[64, 65].

Psychiatric disorders were common comorbidities among HCV-infected individuals in the interferon-era[13, 66], and this trend continues. Almost half of the patients reported a psychiatric diagnosis (44%), utilizing mental health treatment or services (41%), or psychiatric medication use (50%), with 30% currently on psychiatric medications. Eighteen percent (n = 287) of the cohort reported a past history of inpatient psychiatric hospitalizations, a proxy for severe psychopathology. The prevalence of these mental health indicators far exceeds the prevalence in the general US population[67]. A total of 36% of patients were still using alcohol at time of treatment initiation and at least 20% had used street drugs and 9% reported using prescription medications in an abusive fashion in the previous year. While these data are alarming, it is likely that patients under-reported these behaviors due to social desirability and fear of treatment being rescinded. In this study, we attempted to mitigate under-reporting by highlighting the confidential nature of data collection and separation of research team and clinical staff. The high prevalence of psychiatric disorders and moderate drug and alcohol use has clinical relevance for all practitioners, as these patients will be at risk for long-term poor health outcomes, beyond liver disease.

The vast majority of patients suffer from multiple chronic health conditions. Patients reported up to 15 comorbid conditions and an average of four. A greater number of comorbidities was found for patients who are disabled and older. Many of the prevalent medical conditions could represent biologically plausible inflammatory conditions, or extrahepatic disorders (EHDs) related to HCV, such as diabetes[68, 69]. Recent literature suggests that EHDs are underestimated because they are non-specific, but may compromise overall health outcomes and result in an estimated economic burden of $1.5 billion per year[70]. The diagnostic guidelines published by the International Study Group of Extrahepatic Manifestations Related to HCV recommend evaluating all patients with HCV for potential EHDs to ensure that the entire spectrum of HCV-related disorders are identified and properly treated[4]. Indeed, patients with all stages of liver disease should be treated for HCV and not delayed until fibrosis advances, as studies indicate that clinical and economic burden of EHDs can be reduced through viral eradication[70, 71].

Like comorbidities, many, but not all patients reported experiencing specific symptoms. For each of the specific neuropsychiatric, somatic and GI symptoms, a proportion of patients did not experience that particular symptom (e.g., 50-60% had no GI symptoms), but the majority reported mild to severe symptoms. For instance, fatigue and sleep disturbance were ubiquitous problems in the entire cohort. With regard to the neuropsychiatric cluster (depression, anxiety, anger, cognitive concerns), about 60-80% endorsed mild to severe symptoms, with severe neuropsychiatric symptoms experienced by patients who were disabled, unemployed, and on psychiatric medications, among other factors. Over 60% of patients endorsed pain interference and 30% endorsed joint/muscle pain, which is consistent with previous reports of pain disorders in patients with chronic HCV[6, 15, 72–74]. Over 90% of patients reporting some level of sleep disturbance; 31% reported a sleep disorder. Estimates from the literature suggest that up to 60% of patients with chronic HCV may have sleep problems, including restless leg syndrome, although confounding effects of other comorbidities has been difficult to tease apart[6, 75]. Disability, unemployment, and psychiatric medication use were the prominent predictors of the pain, sleep interference, and fatigue cluster. Finally, 40-50% of patients endorsed GI symptoms (nausea/vomiting, abdominal pain, diarrhea) and similarly, disability, unemployment and psychiatric markers including psychiatric medication use were associated with more severe GI symptoms.

A comprehensive analysis of symptom prevalence in the chronic HCV population has not been conducted since the work of Lang et al. in 2006[6]; therefore, this study provides a contemporary perspective of the most prevalent, severe, and distressing symptoms in the chronic HCV population. The majority of symptoms endorsed by at least 20% of patients in this cohort overlap considerably with Lang et al.’s findings, validating our results. We also identified new symptoms (e.g., numbness and tingling in hands and feet, dry mouth, cough, feeling bloated, shortness of breath, and lack of sexual interest) that were prevalent and require further investigation. This information can help clinicians identify and better care for these patients and observe changes in symptoms after viral eradication.

The functional well-being of the current cohort was comparable to that obtained from the original HCV-PRO analysis embedded in a Phase II clinical trial[40]. Items endorsed on the HCV-PRO most often included “Having Hepatitis C was very stressful to me”, “I had difficulty sleeping or slept too much”, and “I needed to pace myself to finish what I had planned”. HCV-specific functional well-being was worse in patients with advanced liver disease (APRI > 2), and those disabled, unemployed and on psychiatric medications. The HCV-PRO is a contemporary disease-specific instrument developed with rigorous methods and we would recommend it for use in future HCV investigations; unfortunately, no other studies have been published to date, therefore cross-study comparisons are unable to be conducted.

Not surprisingly, we found that laboratory biomarkers and markers of liver disease severity were not strong predictors of the PROs investigated in this study. We observed differences in the unadjusted means between high and low MELD subgroups by 2 to 7 points across the PROs, reflecting possible clinically significant differences in patients with advanced cirrhosis on symptoms such as sleep disturbance and the GI symptom cluster. However, MELD was not a strong predictor in confirmatory models. There were small differences between patients with high versus low APRI, FIB-4, albumin and creatinine, but only APRI was validated as a predictor of functional well-being in multivariable models. Our findings are consistent with a large body of HrQOL literature from the interferon treatment era that found negligible associations between liver disease parameters and PROs once other variables were accounted for[15, 76–81]. Most of these studies have found psychiatric and/or medical conditions to be the strongest predictors of PROs in multivariate models, especially depression and pain-related conditions[15, 80–82]. Many clinical researchers have speculated that HCV exerts a deleterious effect on PROs through medical and psychiatric comorbidities, which could represent underlying HCV-induced EHDs [15, 80, 81]. These speculations require further investigation.

Limitations of this study should be noted, particularly regarding the generalizability of these results. Our findings may not apply to other subpopulations or clinical settings, including younger people who inject drugs, persons receiving medication-assisted treatment for opioid use disorders with methadone or buprenorphine/naloxone, Veterans, and those incarcerated. These subgroups may represent the primary treatment groups in years to come if the opioid epidemic is not controlled[83]. We found no substantial differences between patients enrolled and not enrolled in this study, therefore we can presume generalizability to similar patients treated in similar settings. Social desirability and response bias issues could have occurred, especially in response to alcohol and drug use questions. However, it’s important to appreciate the necessity of capturing data directly from patients when they are uniquely positioned to assess an outcome (i.e., pain), as research shows that clinicians tend to under-report the frequency and severity of patients’ symptoms[84]. Finally, minimal clinical data were extracted from medical records, thus we are unable to verify patient-reporting of medical comorbidities, psychiatric medications and other clinical information.

Future research should build upon our observations to directly inform clinical knowledge and decision-making for various stakeholders (patients, clinicians, policy makers). There is a critical need to tease apart and elucidate the causal pathways that may influence health outcomes in the HCV population (i.e., psychiatric illness, EHDs, number of medical comorbidities, liver disease, symptoms); this may be achieved through sophisticated path analyses and structural equation modeling[85]. We identified a new predictor of worse PROs, namely employment status, which should be included in future predictive models of HCV-related PROs. Health services research is needed to examine the most effective models to integrate healthcare for liver disease, mental health, and addiction[21, 55–57, 60–63]. More studies derived from real world clinical settings, outside of clinical trials, need to examine changes in overall symptom burden and specific symptom clusters over time, during DAA therapy, after virologic cure, and during long-term follow-up. Comparisons between the DAA regimens and how they affect PROs would provide stakeholders with information to guide treatment decision-making. Importantly, it remains unclear if overall symptom burden lessens after virologic cure or if only specific symptom clusters improve. Therefore, the exploratory work herein to identify and validate the most robust predictors of PROs is fundamental to future PRO research in chronic HCV.

Our findings have a number of strengths and highlight clinical relevance for multiple stakeholders. This study is unique and notably the largest investigation of PROs in the current population of HCV patients seeking DAA therapy outside of clinical trials. We have identified some of the most challenging sociodemographic features, common comorbidities, potentially EHDs, and troublesome symptoms experienced by HCV-infected individuals. More black Americans have been recruited than any previously conducted clinical trial and thus provided an opportunity for a robust evaluation. Interestingly, Black patients reported the lowest levels of symptom burden and neuropsychiatric symptoms. Consistent with many prior studies, we found that laboratory biomarkers and liver disease severity were not helpful in identifying patients with diminished PROs, such as HrQOL, when other variables were accounted for[80–82]. It is also noteworthy that PROP UP is a highly patient-centered study. Patients affected by HCV were engaged since its inception and helped to identify the study outcomes and select PRO measures to ensure the study findings were salient and meaningful to people with HCV. Finally, given the large sample size, we were able to conduct cross-validation analyses to rigorously identify and confirm predictors, essentially conducting two separate identical studies, and then followed up with sensitivity analyses to confirm that perturbations in modeling, missing data, and assumptions did not alter our findings. These methods serve to instill a higher level of trust in the conclusions drawn from this study. Consequently, we hope the comprehensive data provided herein serves as reference guide for future investigations of PROs in the chronic HCV patient population.

Supporting information

Acknowledgments

The authors would like to acknowledge the contributions of the following people: Alan Franciscus of www.HCVadvocate.org, Anquenette Sloan, Summer Wadsworth-Delciotto, Scott Kixmiller, Larry Huston, Finton Brown (UNC Patient Engagement Group); Virginia Sharpless, Ken Berquist, Herleesha Anderson, Courtenay Pierce, Jenn Barr, Bryonna Jackson, Jane Giang, Jama Darling, Paul Hayashi, Steven Zacks, A. Sidney Barritt IV, Scott Elliot, Dawn Harrison, Danielle Cardona (University of North Carolina); Patrick Horne (University of Florida); Kelly Borges, Danielle Ciuffetelli (University of Pennsylvania); Chrissy Ammons, Kathleen Genther, Jessica Mason (Virginia Commonwealth University); Lelani Fetrow, Vicki Shah (Rush University); Theresa Cattoor, Alisha McLendon (Saint Louis University); Mariechristi Candido, Sophia Zaragoza, Sandeep Dhaliwal, Patricia Poole, Rebecca Hluahanich, Kathleen Haight, (University of California, Davis); Andrea Gajos, Elizabeth Wu, Carrie Bergmans (University of Michigan); AnnMarie Liapakis, MD, Kristine Drozd, Carol Eggers, Hong Chau and Claudia Bertuccio (Yale University). William Harlan, Roberta Golden, Kylee Diaz(Asheville Gastroenterology Associates), William King, Megan Marles (Wilmington Gastroenterology Associates).