Five Laboratory Tests Predict Patient Risk and Treatment Response in Hepatitis C: Veterans Affairs Data from 1999-2010

Background and Aims: New Hepatitis C [HCV] drugs are expensive but highly effective but have created an overwhelming ‘cash flow’ problem for payers facing a large bolus of infected patients. Health plans are developing ‘watchful waiting’ strategies to safely defer treatment for low risk patients. This study identifies five laboratory tests which predict increasing risk of liver related events and documents that delaying treatment until after an abnormal lab test diminished effectiveness of treatment with interferon alpha plus ribavirin. Methods: Patients from the Veterans Administration’s clinical registry of HCV patients [1999-2010] were screened for a detectable viral load at baseline and a recorded baseline genotype. The primary outcomes were time to death and time to time to a composite clinical event. Cox proportional hazards models were estimated with time dependent independent variables for initial treatment and first abnormal laboratory test. Results: 128,769 patients met all inclusion criteria. Abnormal values for five laboratory were associated with increased risk for the composite outcome/death: 1.35/1.84 for the AST/ALT ratio > 1; 2.35/5.01 for albumin < 3 g/Dl; 1.58/1.15 for GGT > 195 IU/L; 3.85/1.55 for platelet count < 100 k/mm and 4.48/2.39 for alpha fetoprotein > 144 ng/mL. Delaying drug therapy until after an abnormal lab test significantly reduced treatment effectiveness. Discussion: Five tests predict liver complications for HCV patients which can be used to develop “watchful waiting” protocols which monitor untreated HCV patients over time and defer access to expensive new drug regimens to those patients most at risk adverse liver-related events.


Introduction
Hepatitis C (HCV) affects approximately 130-170 million people worldwide [1,2] and an estimated 3.2 million people in the U.S. [3] HCV patients are at risk for developing progressive liver disease including cirrhosis, liver failure and hepatocellular carcinoma (HCC). [1,2,[4][5][6] However, many patients with chronic HCV infections may fail to develop significant symptoms over their lifetime. Progression to cirrhosis is estimated to be about 5-20% after 20 years of infection. [7] It is estimated that 13.1% of patients with HCV in 2005 will die of liver-related causes by 2030 [8], increasing to 36.8% by 2060. [9] While HCV is considered cured if the patient achieves sustained viral response (SVR) [10], achieving SVR using older therapies [pegylated interferon-alpha and ribavirin] has been daunting. Standard therapy requires 24 to 48 weeks of weekly injections and is associated with significant side effect burden. Standard therapy achieves SVR rates ranging between 14-24% for HCV genotype 1 and 37-52% for genotypes 2 or 3. [11][12][13] More recent "triple" therapy, achieved by adding a protease inhibitor to pegylated interferon-alpha and ribavirin, is associated with a significant increase in SVR rates and an increase in frequency and severity of adverse side effects such as anemia, neutropenia, thrombocytopenia, rash, and gastrointestinal events. [14] Other reports suggest that 'triple therapy" may increase SVR rates while reducing the side effects that limit tolerability. [15][16][17][18][19][20][21] It is not surprising that clinicians patients faced the twin problems of patient reluctance to initiate therapy and non-adherence once treatment is initiated.
The FDA has approved two new therapies [Fall 2014] which have radically changed the decision by patients and physicians when to initiate treatment by radically increasing expected SVR rates and shortening the duration of therapy. However, the costs of these therapies have created a need to ration drug using 'watchful waiting'/deferred treatment strategies under which clinicians and patients monitor the patient's risk of disease progression over time and initiate treatment before its effectiveness is adversely impacted.
The objective of this research is twofold. First, we use a large historical cohort of HCV patients from to document which specific abnormal laboratory tests predict an increased risk of liver-related liver events, including death. We then investigate whether or not a delay in initiating standard treatment until after one or more of these laboratory tests become abnormal degrades treatment effectiveness relative to 'early' treatment.

Data
The data used in this study were derived from the Veterans Administration [VA] clinical case registry (CCR) for HCV infected patients. Potential HCV patients were identified by the presence of an HCV-related ICD-9 diagnosis code or a positive viral load lab test. Local CCR coordinators then manually confirm or reject the patient for inclusion in the CCR and all historical data from the patient's electronic medical record (EMR) were added to the CCR. The VA EMR system was fully implemented in 1999 and provides data on patient demographic and clinical characteristics, diagnoses, laboratory test results and prescription and medical care utilization for over a decade. [22] Data period for this study cover 1999-2010.
A patient-level analytic database was created consisting of summary variables for each month before and after the patient's date of diagnosis [CCR enrollment date]. The post-index period for any individual patient could range up to 10 years depending on their index date. The following summary data were created:  [1999][2000][2001][2002][2003][2004][2005][2006][2007][2008][2009][2010] and across multiple VA sites that reported data into the CCR. This selection criteria errored on the side of including patients whose viral load was reported as 'undetectable' at a higher minimum value [e.g., < 600 IU/ml] using older testing methods.

Primary and Secondary Outcomes
Individuals infected with HCV are at risk for progressive liver disease and related complications such as cirrhosis, liver failure and hepatocellular carcinoma (HCC) which frequently require hospitalization and are at increased the risk of death. [1,2,[4][5][6] The primary outcomes for this analysis are a composite of observed liver-related, clinical complications [compensated cirrhosis, decompensated cirrhosis, HCC or a liver-related hospitalization] or all-cause mortality. To ensure the clinical complications were due to chronic HCV infection, any composite clinical outcome occurring within a one-year "washout" period following the index date were not counted. The secondary outcomes included the individual elements of the clinical composite analyzed individually.
Monthly dichotomous variables were created for the outcomes of the study based on recorded diagnostic codes [e.g., diagnosis of cirrhosis, etc.] and selected CPT-4 codes included in data from hospital admission data and outpatient services. Liver-related hospitalization events were defined as the first occurrence of a diagnosis for complications due to HCV (see Appendix 1) as the primary diagnosis in inpatient visits. Compensated cirrhosis and HCC outcomes were compiled by searching the inpatient, outpatient and problem lists for matching ICD-9 codes [571. 5 [23] or an Ishak fibrosis stage of F4-F6. [24] Vallet-Pichard, et al. also found the FIB-4 to have a positive predictive value to confirm the existence of significant fibrosis of 82.1% in a HCV infected cohort. [23] Statistical Methods The time-to-event variables for primary and secondary 12 Five Laboratory Tests Predict Patient Risk and Treatment Response in Hepatitis C: Veterans Affairs Data from 1999-2010 outcomes were analyzed using Cox proportional hazards. [25] Time-related independent variables were created for several risk factors to take advantage of the longitudinal nature of the data. The time-related laboratory test risk factors assessed whether or not the laboratory test values were abnormal immediately prior to the event date. The time-related treatment variable was included to assess whether or not treatment was initiated before or after the occurrence of each liver-related outcome variable. The time at which treatment is initiated is endogenous over time and correlated with observed and unobservable patient factors that are also correlated with patient outcomes. Therefore, the treatment effects measured in this analysis may be biased to the extent that important but unobserved patient characteristics are not included in the statistical models. However, the objective of this research is to test whether or not initiating treatment after one or more of the patient's laboratory value becomes abnormal rather than develop an unbiased estimate of treatment effects. While complicated, the methods used are at the heart of the analysis which ultimately provides information on what do clinicians watch to determine whether or not treatment is needed, and whether or not it is possible to wait too long, thus diminishing the effectiveness of treatment at reducing future risk. The answer to this question will be measured by the estimated coefficient for the interaction term between the time dependent dichotomous variables for treatment and the existence of one or more abnormal lab test.
A total of 22 individual clinical laboratory tests were entered into the analyses of each adverse liver event and death in order to determine which abnormal laboratory tests were predictive of these events. Laboratory test result data were entered as time-to-abnormal test result and the specific cut-points for defining an abnormal value can be found in Table 2. These cut-off values were set by the authors at a level high enough to generate wide agreement that the recorded lab value was abnormal.
All study patients were required to have a detectable viral load in order to focus the analysis on patients who could benefit from treatment. Separate Cox models were estimated for each primary and secondary outcome under study. Each analysis began by estimating a Cox model with all possible risk factors included in the list of independent variables, including the complete list of 22 common laboratory tests thought to be related to liver related adverse events. Since models of event risk should be parsimonious to be useful in clinical practice, the results for these all-inclusive models were reviewed and non-significant factors [p>0.20] were dropped from subsequent model estimates. Age, gender, HCV genotype, and race were included in all models regardless of their statistical significance. Race and ethnicity were initially entered as separately coded categories. However, the significant correlation between race and ethnicity in this VA sample resulted in ethnicity being dropped from the final model specification. The dummy variables denoting a diabetes diagnosis at baseline and any hospital admission in the 6 months prior to the patient's index date were included in final model specifications based on statistical significance.
Once we documented which abnormal laboratory tests were predictive of future liver events, we tested if treatment effectiveness was dependent on the timing of treatment either before or after any one lab value becomes abnormal. This was accomplished by entering both the time dependent treatment variable and its interaction term with a time dependent variable denoting whether or not any abnormal laboratory test preceded treatment. This interaction variable measures the extent to which the effect of treatment initiated after an abnormal laboratory test is statistically different from the impact of 'early' treatment. The

Results
Descriptive Statistics

Predictive Laboratory Tests
The laboratory data in Table 2 document the deteriorating health status of the study HCV population over time. The proportion of patients with abnormal values is higher in the post-index period for all laboratory tests with the exception of HbA1c for diabetes and for viral load. The laboratory tests found to be correlated with the risk of long-term liver complications and death are denoted by a bold font in Table  2.
The parsimonious models for each of our primary and secondary outcomes are displayed in Table 3 [Primary Outcomes] and Table 4 [Secondary Outcomes]. The 'side-by-side' arrangement of the models makes it easy to evaluate the extent to which risk factors are shared across target outcomes, and the relative strength of risk factors across outcomes.  a. Lab tests that were included in the initial model specifications as potential risk factors but which never achieved statistical significance in any risk prediction model include: hemoglobin, white blood cell count, potassium *p<0.05; **p<0.01; ***p<0.0001 a. Lab tests that were included in the initial model specifications as potential risk factors but which never achieved statistical significance in any risk prediction model include: hemoglobin, white blood cell count, potassium *p<0.05; **p<0.01; ***p<0.0001 All four clinical outcomes share common results for the five laboratory tests (ALT/AST ratio, albumin, GGT, alpha fetoprotein, platelets) that increased the risk of the composite clinical outcome and death [ Table 4]. Several additional laboratory tests have crept into the individual models especially for cirrhosis, in which serum iron, ferritin and total iron binding capacity [TIBC] are predictive of higher risk, but an abnormal red blood cell count and blood urea nitrogen are predictive of lower risk. Other 'protective' abnormal lab values are blood urea nitrogen [BUN] for HCC and abnormal neutrophil counts for decompensated cirrhosis. Abnormally high alkaline phosphatase [ALP] is predictive of increased risk of liver-related hospitalizations and HCC.

Delaying Treatment
Estimates of the impact of initiating treatment before and after the emergence of an abnormal laboratory tests are presented in [ Table 5]. The second column of Table 5 repeats the average results of drug therapy effectiveness found in Tables 3 and 4 to assist in interpreting results. The last two columns of Table 5 report our estimates of the effectiveness of drug therapy started 'early' [before any abnormal laboratory value is reported] or 'late' [following an abnormal laboratory test]. There are two tests of significance provided for the estimated effectiveness of 'late' treatment. The first measure of statistical significance denoted by '*' footnotes designates whether or not the estimated effect is different from the null hypothesis. The second statistical test, designated by 'a' indicates whether or not the estimated effects for early and late initiation of drug therapy are statistictically different. That is, did delaying the initiation of treatment adversely impact the effectiveness of initiating treatment?
There are two basic results of significance. First, the initiation of treatment prior to any laboratory test becoming abnormal is uniformly protective, but limited. Early . The limited impact of initiating treatment is not surprising given that only 16.4% of patients who initiate treatment achieve an undetectable viral load. Second, initiating standard HCV treatment after one or more laboratory test becomes abnormal never has a significant impact on the risk of these events. That is, delaying standard treatment until the patient exceeds any one of the lab test upper bounds listed in Table 2 significantly degrades the effectiveness of treatment.

Discussion
Tools are needed with which to manage newly approved drug therapies. Our results also suggest that delaying therapy until the patient's lab profile becomes 'abnormal' could compromise patient outcomes. Our results demonstrating the limited effectiveness of standard therapy are also consistent with earlier research using VA data [11] and other 'real world' treatment settings. [12,13] Treatment rates in our VA sample using standard therapy are only 24.3% and few treated patients [16.4%] achieved at least one undetectable level of viral load test [4% of the total HCV study population]. It is no surprise that asymptomatic HCV patients have gambled in the past that they will not develop HCV complications in order to avoid the treatment burden associated with standard therapy.
New treatment options have reversed the dynamics of the treatment decision. Patients are now demanding treatment with expensive, efficacious and less burdensome therapies while payers seek to ration treatment to high risk patients. A cost-containment strategy of reserving treatment for patients 'abnormal' lab tests cannot be implemented because the thresholds used here to define abnormal lab tests were set high to avoid controversy about whether or not the test was truly abnormal. More research is needed to set the screening values for each of the 5 test in this analysis at a level high enough to predict increased event risk but low enough to avoid inappropriate delays in treatment that impact the effectiveness of treatment.
Previous research has found similar relationships between the risk of liver events and abnormal laboratory test results based on relatively small samples of patients with disease progression. Lee, et al. [27] found that elevated ALT levels were associated with an increased risk of HCC in Taiwan. Hu, et al. [28] found that elevated AST and alpha-feta protein [AFP] were associated with stages 3-4 fibrosis in Taiwan patients. In a Japanese study of 459 patients, Imazeki, et al. [29] found that increasing fibrotic stage was highly correlated with the risk of all-cause mortality, but baseline ALT, AST, albumin and platelet counts were not statistically significant when fibrosis stage was entered as an explanatory variable. In a second small Japanese study [n=345], Shiratori, et al. [30] found elevated albumin to be protective for HCC and elevated AST to increase the risk of death. In a small study in Spain [N=568]. Fernandez-Rodriguez, et al. [31] found that baseline albumin [<3.9 g per 100 ml] was associated with a three-fold increase in the risk of long-term liver complications. In a study of treated patients, van der Meer, et al. found that an increasing AST/ALT ratio was associated with increased risk of mortality. [32] A study that used VA data found that abnormal laboratory values for albumin, AST, AST/ALT ratio, creatinine clearance, and sodium were associated with mortality in genotype 1 patients. [33] Finally, the HALT-C study found that changes from baseline for platelets, AST/ALT ratio, bilirubin and albumin were predictive of decompensated cirrhosis and liver-related death or liver transplantation. [34] There are several additional and important limitations in our study. First, most HCV cohorts from the US would not match well with our study sample which consist almost exclusively of men. Second, treatment for HCV patients in the VA health care system may vary from treatment patterns found in the civilian health care system. However, the Veteran Administration is the largest provider of care to chronically HCV-infected patients in the US. [35] The asymptomatic nature of HCV and the retrospective design of this study make it likely that the index date identified here will not correspond to the actual date of HCV infection. However, this uncertainty regarding the actual date of infection is common in clinical practice. Most patients with hepatitis C are unaware of their infection until their exposure to HCV is detected by routine screening of blood donations or their health status deteriorates due to the infection.
The changing sensitivity of HCV viral load tests also presented a challenge in defining an 'undetectable' viral load. While many older tests have a lower threshold of 600, newer tests are sensitive up to 10 IU/mL. We chose to apply a cutoff of 25 as the limit of 'detectable' for all patients to be consistent, but it is highly likely that some patients with older labs will be mis-categorized due to the low sensitivity of the tests in use at that time.
Finally, our study does not capture medical care outside the VA system, such as the Medicare program. If patients received treatment for complications of liver disease outside the VA, these data may not have been captured and time to event data will be accurate or missing. However, we believe that most of the care for HCV patients was provided within VA system due to the complete coverage for prescription medications and the very high cost of HCV related medications.