Using the Electronic Health Record to Characterize the Hepatitis C Virus Care Cascade

Abstract

Objectives

Chronic hepatitis C virus (HCV) infection is one of the main causes of hepatocellular carcinoma. Before initiating a multilevel HCV screening intervention, we sought to (1) describe concordance between the electronic health record (EHR) data warehouse and manual medical record review in recording aspects of HCV testing and treatment and (2) estimate the percentage of patients with chronic HCV infection who initiated and completed HCV treatment using manual medical record review.

Methods

We examined the medical records for 177 patients (100 randomly selected patients born during 1945-1965 without evidence of HCV testing and 77 adult patients of any birth cohort who had completed HCV testing) with a primary care or relevant specialist visit at an academic health care system in Tampa, Florida, from 2015 through 2018. We used the Cohen κ coefficient to examine the degree of concordance between the searchable data warehouse and the medical record review abstractions. Descriptive statistics characterized referral to and receipt of treatment among patients with chronic HCV infection from medical record review.

Results

We found generally good concordance between the data warehouse abstraction and medical record review for HCV testing data (κ ranged from 0.66 to 0.87). However, the data warehouse failed to capture data on HCV treatment variables. According to medical record review, 28 patients had chronic HCV infection; 16 patients were prescribed treatment, 14 initiated treatment, and 9 achieved and had a reported posttreatment undetected HCV viral load.

Conclusions

Using data warehouse data provides generally reliable HCV testing information. However, without the use of natural language processing and purposeful EHR design, manual medical record reviews will likely be required to characterize treatment initiation and completion.

Liver cancer incidence and mortality in the United States have increased in recent years. ^1,2 Chronic hepatitis C virus (HCV) infection is one of the main causes of the most common type of liver cancer, hepatocellular carcinoma. ^3,4 HCV infection is often asymptomatic, and fewer than half of people infected with HCV are aware of their infection. ⁵ HCV screening rates are low across adult birth cohorts (from 4.3% among people born before 1945 to 15.7% among people born during 1966-1985 in 2016). ⁶ Even in health care systems with high screening rates, many patients may “fall out” or “drop off” during progression along the cascade of care, from screening through treatment. ^7,8

Interventions are needed to increase HCV screening, linkage to care, and treatment completion. However, before implementing such interventions in a health care system, assessing current HCV testing and treatment rates is necessary to monitor improvements. The electronic health record (EHR) searchable data warehouse will likely be an important tool in assessing rates at which patients progress through the care cascade; most hospital systems have at least a basic EHR. ⁹ The data warehouse may include structured clinical data (eg, laboratory, pharmacy, radiology) and administrative data (eg, demographic information, billing codes). ^{10 -12} To be included in the data warehouse, data must be entered into discrete data/variable fields. However, most EHR data are unstructured (ie, free text). ¹³ Unstructured data include narratives dictated in progress notes (eg, information provided verbally by the patient to the health care provider), typed notes from health care providers/staff members (eg, notes about patient factors that might explain why treatment was not initiated or completed), and documents from other health care sources. It is unknown how easily HCV-related testing and treatment orders and results are accessed from the data warehouse, which could lead to inaccurate estimates of diagnosis and treatment in a patient population ¹⁴ and an inaccurate representation of the care cascade. If data warehouses do not capture this information, a more resource-intensive, manual medical record review of notes and scanned documents may be necessary to obtain this information.

Before initiating a multilevel HCV screening intervention (ie, clinical decision support through the EHR and health care provider education) in an academic health care system, we assessed the completeness and accuracy of HCV screening and treatment data obtained from the data warehouse by comparing them with data from a manual medical record review. Because the data warehouse has only structured data and most EHR data are unstructured, ¹³ we first compared the 2 data sources for concordance on HCV care cascade outcomes. We also described the percentage of patients with chronic HCV infection who initiated and completed treatment via a manual medical record review. The following research questions guided the study:

1. What is the degree of concordance between the data warehouse and medical record review on whether an HCV screening (antibody) test was (1) ordered and (2) completed?

2. Among patients with a positive antibody test result, what is the degree of concordance between the data warehouse and medical record review with regard to (1) confirmatory/diagnostic test (RNA test) ordered, (2) diagnostic test completed, (3) genotype reported, and (4) viral load reported?

3. Among patients with chronic HCV infection, what is the degree of concordance between the data warehouse and medical record review with regard to HCV treatment being (1) ordered, (2) initiated, and (3) completed?

4. On the basis of manual medical record review, what percentage of patients with chronic HCV infection initiate and complete HCV treatment?

Methods

Under a separate protocol to examine HCV testing rates for all patients aged ≥18 years during August 1, 2015–July 31, 2017, a business systems analyst (W.S.) extracted the EHR data from the data warehouse. Methods are described elsewhere. ¹⁵ Briefly, in that retrospective study, we used the data warehouse to identify patients aged ≥18 years seen at University of South Florida (USF) Health, a large academic medical health care system. The previous study examined the prevalence of HCV screening orders by age cohort and examined factors associated with having an HCV screening ordered among patients at average risk of HCV infection. ¹⁵ We included patients who had at least 1 visit with either a primary care provider (eg, family medicine, internal medicine) or a relevant specialist (eg, gastroenterologist, infectious disease) for outpatient care during that period. We excluded patients who did not have a relevant visit (eg, podiatry). The final data set included 65 114 patients. Hereinafter, we refer to this data set as the data warehouse data.

The previous study found that the frequency with which HCV screening orders were placed increased among patients born before 1945 and during 1945-1965, did not change among patients born during 1966-1985, and declined among patients born after 1985. ¹⁵ Among patients born during 1945-1965, multiple sociodemographic and health care factors were associated with having an HCV screening test ordered. ¹⁵ Screening completion was not examined. In the current study, we examined screening orders, screening completion, RNA testing, and treatment initiation and completion, using both the data warehouse and manual medical record review. A biostatistician (R.R.R.) programmed a search function to identify HCV-related procedure orders (eg, HCV antibody test, HCV reflex to quantitative, HCV RNA test) and results in the database.

To address research questions 1-3, we randomly selected 100 patients born during 1945-1965 without evidence of HCV antibody testing using a random number generation script from the data warehouse data set for a medical record review to evaluate the validity of the data warehouse’s showing receipt of no HCV testing. To assess treatment and follow-up (research question 4), we had aimed to select 100 HCV antibody–positive patients born during 1945-1965 for manual medical record review. However, because the data warehouse data indicated that only 77 patients aged ≥18 years from any birth cohort had an HCV antibody or RNA test result during the study period, we included all 77 patients in the manual medical record review. For the fourth research question, we conducted a manual medical record review. To better assess the proportion of patients completing the care cascade, we extended the study period for the medical record review by 1 year (to July 31, 2018) for 12 patients diagnosed with HCV infection near the end of the original study period but unable to complete care before the study end date. The USF Institutional Review Board approved both study protocols and granted a waiver of informed consent.

Strategic Planning, Business Development & Analytics at USF Health acted as an honest broker by generating random subject identification numbers for the study and linking the identification numbers to medical records for the medical record review. Employees of the health care system (a physician [R.G.R.] and 3 trained medical assistants) abstracted data from patients’ medical records, including searching media files where scanned documents may be saved. Initially, all reviewers reviewed the same 5 medical records to ensure agreement on major data elements, with 100% agreement. The medical record reviewers noted any HCV-related testing and treatment in an abstraction form developed in collaboration with clinical colleagues. The medical record abstraction form (supplemental material available upon request) captured data on HCV screening (eg, antibody test ordered [yes/no], results), liver diagnoses (eg, HCV, liver cancer), screening indicators (eg, birth cohort), antibody-positive test (yes/no), referral to a specialist (yes/no), confirmatory testing (eg, RNA test ordered [yes/no], confirmatory test completion [yes/no], genotype ordered [yes/no], genotype completion [yes/no]), testing/follow-up (eg, imaging), treatment prescribed (eg, medication), treatment receipt (eg, posttreatment viral load assessment, comments), and behavioral factors (eg, drug use). Generic and trade names for all possible HCV treatments were searched in a field called “Ordered_Medication.” Medical record reviewers entered data into a medical record review–specific database (separate from the data warehouse data).

Statistical Analyses

To address the first 3 research questions, we used the Cohen κ coefficient to examine the degree of concordance between the data warehouse and medical record review for each variable that had a yes/no response, using the following κ levels and guidance for strength of agreement: <0 = poor, 0-0.20 = slight, 0.21-0.40 = fair, 0.41-0.60 = moderate, 0.61-0.80 = substantial, and 0.81-1.00 = almost perfect. ¹⁶ For genotype and viral load reporting, we used percentages to characterize concordance between the data warehouse and medical record review. To address the final research question on treatment initiation and completion, we used percentages. We defined treatment completion as receiving HCV treatment with subsequent posttreatment viral load testing.

Results

Most of the 177 patients in the total sample were female (58.2%), non-Hispanic White (58.2%), and had private health insurance (57.6%) (Table 1). The mean age of patients was 53.8 years. Among patients completing HCV antibody testing, most were female (60.0%), non-Hispanic White (46.3%) or non-Hispanic Black (20.0%), and had private health insurance (67.5%). The mean age of people who completed HCV antibody testing was 45.0 years. Concordance between the data warehouse and medical record review was substantial on test ordering (κ = 0.77), test completion (κ = 0.78), and receipt of positive antibody result (κ = 0.66) (Table 2). In addition to the 63 patients who had an indication of having had an HCV antibody test ordered in the data warehouse, we found another 18 patients with no evidence of HCV antibody testing ordered in the data warehouse to have had an HCV antibody test ordered in the manual medical record review (for a total of 81 patients with HCV antibody testing ordered).

Table 1.

Characteristics of University of South Florida (USF) Health patients aged ≥18 years across the hepatitis C virus (HCV) treatment cascade of care, August 1, 2015, through July 31, 2018, Tampa, Florida^{a ,b,c}

Characteristics	Total sample for data warehouse and medical record review (N = 177)	Antibody screening completion (n = 80) d	Patients with chronic HCV infection e (n = 28)	Patients receiving treatment with posttreatment viral load testing (n = 9)
Age, y
Mean (SD)	53.8 (14.2)	45.0 (16.2)	55.6 (12.5)	61.4 (6.2)
Median (range)	57 (19-72)	48 (19-72)	59 (24-68)	63 (51-69)
Sex
Female	103 (58.2)	48 (60.0)	14 (50.0)	5 (55.6)
Male	74 (41.8)	32 (40.0)	14 (50.0)	4 (44.4)
Race/ethnicity
Hispanic	23 (13.0)	11 (13.8)	3 (10.7)	1 (11.1)
Non-Hispanic White	103 (58.2)	37 (46.3)	16 (57.1)	5 (55.6)
Non-Hispanic Black	22 (12.4)	16 (20.0)	5 (17.9)	2 (22.2)
Non-Hispanic Asian	7 (4.0)	3 (3.8)	1 (3.6)	0
Non-Hispanic other	22 (12.4)	13 (16.3)	3 (10.7)	1 (11.1)
Primary language
English	168 (94.9)	75 (93.8)	27 (96.4)	9 (100.0)
Spanish	4 (2.3)	3 (3.8)	0	0
Other	2 (1.1)	2 (2.5)	1 (3.6)	0
Health insurance coverage
Medicaid	13 (7.3)	7 (8.8)	3 (10.7)	1 (11.1)
Medicare	51 (28.8)	15 (18.8)	11 (39.3)	6 (66.7)
Medicare supplement	3 (1.7)	1 (1.3)	1 (3.6)	0
Military	6 (3.4)	3 (3.8)	1 (3.6)	0
Private	102 (57.6)	54 (67.5)	12 (42.9)	2 (22.2)
Other	2 (1.1)	0	0	0
No. of physician visits
Mean (SD)	6.9 (20.0)	8.5 (11.6)	10.6 (11.2)	4.3 (1.9)
Median (range)	4 (1-67)	5 (1-63)	6.5 (1-49)	5 (1-7)

^aData source: USF Health data warehouse abstraction ¹⁵ and medical record review. All data are number (percentage) unless otherwise indicated.

^bTotal sample of 177 patients included 100 patients born during 1945-1965 with no evidence of HCV antibody screening and 77 adult patients (all birth cohorts) with an HCV antibody test result seen at USF Health during the study period.

^cThe initial study period examining HCV testing rates was August 1, 2015–July 31, 2017, but the study period was extended by 1 year for 12 patients diagnosed with HCV infection near the end of the original study period but unable to complete care before the study end date.

^dPatients with evidence of HCV antibody completion included 77 patients with an indication of HCV antibody testing in the data warehouse and 3 patients in the data warehouse who had no evidence of HCV antibody testing but had HCV antibody testing in the manual medical record review.

^ePatients with chronic HCV infection were defined as those with an indication of either a positive RNA test result or HCV genotype recorded in either the data warehouse or medical record review.

Table 2.

Degree of concordance between the data warehouse and manual medical record review abstractions for hepatitis C virus (HCV) screening and confirmatory testing, August 1, 2015, through July 31, 2017, University of South Florida (USF) Health, Tampa, Florida^{a ,b}

HCV testing variable	Outcome reported in data warehouse, no. (%) (N = 177) c	Outcome reported in medical record review, no. (%) (N = 177) c	Cohen κd,e
HCV antibody test ordered	63 (35.6)	81 (45.8)	0.77
HCV antibody test completed	63 (35.6)	80 (45.2)	0.78
Positive antibody test result reported	28 (15.8)	41 (23.2)	0.66
HCV RNA test ordered	38 (21.5)	38 (21.5)	0.87
HCV RNA test completed	37 (20.9)	36 (20.3)	0.84
Positive RNA test result reported	22 (12.4)	25 (14.1)	0.83

^aData source: USF Health data warehouse abstraction ¹⁵ and medical record review.

^bThe study period was extended by 1 year for 12 patients diagnosed with HCV infection near the end of the original study period but unable to complete care before the study end date.

^cA denominator of 177 patients was used for these calculations because some patients had an RNA test ordered/completed but not an antibody test ordered/completed in our data set.

^dCohen κ coefficients suggest almost perfect concordance (defined as κ = 0.81-1.00) for RNA variables examined and substantial concordance (defined as κ = 0.61-0.80) for the antibody tests examined. ¹⁶

^eCohen κ coefficients could not be obtained for the treatment-related variables because the data warehouse failed to capture most of this information.

We found 28 patients with chronic HCV infection identified through subsequent testing (ie, indicated as either a positive RNA test result or HCV genotype) in either the medical record review or data warehouse; 50% were female and 57.1% were non-Hispanic White (Table 1). The mean age of patients with chronic HCV infection was 55.6 years. Concordance between the data warehouse and medical record review on whether confirmatory (HCV RNA) testing was ordered and completed was almost perfect (κ = 0.87 and κ = 0.84, respectively) (Table 2). Concordance between the data warehouse and medical record review on a positive RNA test result was almost perfect (κ = 0.83). Eighteen patients had information on genotype in the data warehouse and medical record, of which 17 were concordant. However, an additional 6 patients had information on genotype in 1 but not both data sources, and 1 patient had a genotype that was determined to be inconclusive. The most common genotype was type 1A. Among patients with chronic HCV infection who had a numeric viral load value reported in the medical record and data warehouse, 78% of the values were concordant.

For treatment-related variables, the data warehouse had information on HCV medication for only 1 patient, and the data warehouse did not indicate HCV treatment initiation or completion data for any patients. Thus, we could not obtain the Cohen κ coefficient for the treatment-related variables. Of 28 patients with evidence of chronic HCV infection, 1 (3.6%) was diagnosed with hepatocellular carcinoma (Table 3). Based on medical record review, 16 (57.1%) patients had evidence of being prescribed HCV treatment, 14 of whom initiated treatment. Nine of the 14 patients who initiated treatment completed treatment, including posttreatment viral load testing with a recorded result, all of whom achieved sustained virologic response (defined as no detectable amount of HCV). The medical record indicated that 3 additional patients (10.7%) initiated and completed HCV treatment and posttreatment viral load testing; however, posttreatment viral load testing results were not recorded in the medical record.

Table 3.

Receipt of care outcomes after hepatitis C virus (HCV) diagnosis, August 1, 2015, through July 31, 2017, University of South Florida (USF) Health, Tampa, Florida (n = 28) ^a

Outcome	No. (%) b
Receipt of care/treatment after HCV diagnosis (n = 28)
Diagnosed with HCC and no indication that HCV treatment was received	1 (3.6)
No indication of HCV treatment/follow-up care or indication of uncertainty in whether received treatment/follow-up care (eg, referred, but unknown whether patient followed through)	5 (17.9)
Indication that follow-up care not received because of patient factors (eg, patient declined care, patient discontinued care, patient not compliant with recommended follow-up, patient did not follow up with additional tests ordered)	6 (21.4)
Indication in medical record that patient received follow-up care and that HCV treatment was prescribed	16 (57.1)
Treatment prescribed (n = 16)
Prescribed treatment, but did not initiate because of lack of health insurance approval	1 (3.6)
Prescribed treatment, but did not initiate because of patient choice	1 (3.6)
Prescribed treatment and indication in medical record that HCV treatment was initiated	14 (50.0)
Treatment initiated (n = 14)
Unknown whether full medication regimen was completed, and no posttreatment viral load testing completed	1 (3.6)
Indication that full HCV treatment regimen received, but posttreatment viral load testing not completed	1 (3.6)
Indication that full HCV treatment regimen received and posttreatment viral load testing occurred, but results unknown	3 (10.7)
Undetected HCV viral load posttreatment (ie, cure achieved)	9 (32.1)

Abbreviation: HCC, hepatocellular carcinoma.

^aData were extracted during the manual medical record review.

^bThe denominator is 28—the number of patients with chronic HCV infection based on an indication of either an RNA-positive test result or HCV genotype recorded in the USF Health data warehouse ¹⁵ or medical record.

Discussion

We found almost perfect concordance ¹⁶ between the data warehouse and medical record review in ordering and completing RNA testing and whether an HCV-positive RNA test result was found, with κ coefficients ranging from 0.83 to 0.87. We found substantial concordance ¹⁶ between the data warehouse and medical record review in HCV antibody test orders, completion, and positivity. High congruence in our study for HCV RNA testing variables is similar to the high congruence found in a study that compared the data warehouse with medical record review for asthma diagnoses in primary care (κ = 0.89). ¹⁷ In our study, concordance between the data warehouse and medical record review for HCV RNA testing variables (almost perfect) was better than for HCV antibody testing variables (substantial). ¹⁶ The instances of incongruence in our study appear to be “misses” in the data warehouse (ie, an absence of data in the data warehouse vs presence of data in the medical record). This finding is consistent with the findings of a Veterans Health Administration study comparing a medical record review performed though an external peer review program with the data warehouse on multiple clinical performance measures for patients with hypertension, diabetes, or heart disease. ¹⁸ In that study, the medical record review provided more complete data than the data warehouse, and κ coefficients were almost perfect (range, 0.86 to 0.99). ¹⁸ Our findings indicate that data warehouse data were largely accurate in describing HCV test ordering and completion for screening and diagnostic testing, which suggest that these data can be used to inform health care system and population-focused HCV testing efforts.

However, our study revealed that the ordering of treatment medications, treatment initiation, and treatment completion were not captured in searchable data warehouse fields. Specifically, information on HCV medication was found for only 1 patient during the data warehouse search, whereas the manual medical record review provided rich and nuanced data about the progression from diagnosis through treatment or, for many patients, the reasons for not initiating or completing treatment. Thus, manual medical record review may be needed to describe the treatment portion of the care cascade. Increasingly, however, opportunities may be available to implement natural language processing algorithms to mine unstructured data on multiple health concerns and in various medical care contexts. ^{19 -24} For example, natural language processing correctly identified 93% of true positive HCV cases in a previous study. ²⁵ Unfortunately, for our study, unstructured data were not searchable with the technology available; furthermore, the data warehouse did not include unstructured data. In addition, health systems might track patients through the care cascade by adding HCV testing and treatment-specific data elements to their EHR. Bringing together a team of health care providers, researchers, and clinical informatics specialists would likely facilitate the design of a user-friendly, comprehensive set of EHR elements, which would produce a high-quality data set to characterize the HCV care cascade. Similarly, a transdisciplinary team will be important in designing natural language processing programs for the HCV care cascade.

Half (14 of 28) of patients with chronic HCV infection initiated treatment, only 9 of whom completed treatment and had posttreatment viral load test results recorded (all of whom demonstrated cure). For several patients for whom treatment was recommended, patient factors (eg, patient declined treatment, did not follow up) were indicated as the reason for treatment nonreceipt. To reduce HCV-related morbidity and mortality, efforts to promote treatment initiation and completion among patients with chronic HCV infection are needed. In addition, the medical records for several patients indicated uncertainty about treatment receipt, which may have resulted from patients receiving care elsewhere. Notes in some patients’ medical records indicated uncertainty about whether care was received, and other medical records lacked such information. Some EHRs have sections where medical information from other institutions might be housed (eg, “care everywhere”), but this information must be accessible to confirm receipt of care. To describe treatment initiation and completion rates and implement effective interventions in a health care system, information on treatment and barriers needs to be recorded. However, high-quality, comprehensive data need to be entered into the EHR to answer questions about barriers to treatment initiation and completion, regardless of whether the data are being searched via manual medical record review, natural language processing, or another method. A transdisciplinary team approach, including physicians, nurses, pharmacists, medical assistants, and laboratory technicians, should be used to ensure that treatment-related variables are consistently entered into the EHR system by those who come into contact with patients throughout the care cascade. Additional insights into effective data tracking and care management across the HCV care cascade might be drawn from the Veterans Health Administration. ^26,27

Limitations

This study had several limitations. First, it was conducted in a single academic medical center with a specific EHR. As such, our findings may not be applicable to other medical settings or systems that do not capture data on HCV screening in the EHR. Second, we conducted our concordance review on a small subset of patients. Third, specific codes were used to identify testing orders and testing completion from the data warehouse, and these codes may have missed some testing occurrences. For example, although we identified misses in the data warehouse, the information may not have been extracted because of being coded incorrectly. Fourth, we focused our examination on a specific date range; if some patients completed HCV testing and treatment outside the study dates, this information would not have been included in the data sets. Finally, previous studies have demonstrated differences across clinic settings and systems in HCV infection rates and the number of patients who complete all steps in the care cascade. ^7,8 Given the small numbers of positive antibody results and patients with chronic HCV infection, we did not examine screening rates and linkage to care by the type of care received or the clinic in which care was received.

Conclusions

Our findings suggest several future directions. One direction is to examine subsets of patients whose data warehouse and medical record review abstractions reveal differences in accuracy (eg, sociodemographic characteristics, department/clinic, reason for visit). In addition, to promote health equity and in light of universal screening recommendations initially for people born during 1945-1965 ^{28 -30} and extended in 2020 to adults aged 18-79, ³¹ it will be important to determine whether certain sociodemographic characteristics and health care experiences are associated with receiving a recommendation for HCV screening and confirmatory testing, completing screening, receiving confirmatory testing, and progressing to treatment. Future research could also examine whether certain sociodemographic characteristics and health care experiences are associated with initiating and completing treatment. Indications of inequity may warrant interventions to facilitate progress through the care cascade. Previous research has suggested that varying rates of testing and linkage to care may occur in a single health care system. ⁸ Using the EHR, health care systems can examine rates of patient receipt of HCV-related care; variations in rates can serve as opportunities for intervention to improve HCV testing and treatment rates.