Costs, Demographics, and Causative Agents in Patients Hospitalized for Drug Induced Liver Injury: Trends in a Large Academic Healthcare System

Introduction

Drug-induced liver injury (DILI) is a complication resulting from medications, herbal supplements, or dietary supplements. Due to its wide spectrum of clinical presentations (from asymptomatic transaminitis to acute liver failure (ALF), its lack of biomarkers, and the need to exclude other causes of transaminitis, it is difficult to diagnose.¹ Population studies have estimated an incidence of 14–19 per 100,000 persons, with a higher incidence of 32.8 per 100,000 persons in hospitalized patients.^1–3 DILI is further characterized as intrinsic versus idiosyncratic.⁴ Intrinsic is attributed to culprit drugs with known liver toxicity in a dose-dependent manner with predictable outcomes, such as acetaminophen. Idiosyncratic refers to drugs that do not have a linear relationship with dosing and are dependent on host susceptibility and variation in presentation. The latter results in delays in diagnosis, owing to the lack of standardized biomarkers and clinical guidelines.

While DILI is considered a rare phenomenon, its consequences can be life threatening. DILI is associated with substantial morbidity and mortality, with a study estimating 9.4% of patients with DILI either dying or requiring a liver transplant within 6 months of DILI onset.⁵ In a prospective cohort study of 17 tertiary centers, Goldberg et al estimated that ALF was secondary to acetaminophen overdose in 39% of cases and idiosyncratic drug reactions in 13% of cases, thus surpassing viral hepatitis as the most apparent cause of ALF.⁶

Common drug classes that cause DILI based on the DILIrank dataset (a ranking of drugs based on their potential of causing DILI) include nonsteroidal anti-inflammatory drugs (NSAIDs), antigout, antimycotics, antineoplastics, psychoanaleptics, immunostimulants, and antivirals.⁷ Interestingly, antibiotics are a common culprit of DILI, but this study did not find it represented a significantly higher hepatotoxic risk as a class due to many antibiotic drugs not being classified as a DILI concern. Other studies, however, have found antimicrobials to be the most common etiology of DILI.^8,9

Overall, DILI can result in serious complications and can arise from a multitude of medications. However, the estimated healthcare expenditures from DILI have not been well described. It has been found that DILI is a major cause of withdrawal of medications from the market.¹⁰ One study estimates that the typical cost for the research and development of a new therapeutic drug now costs $800 million, therefore a withdrawal from the market can be costly.¹¹ The aim of this study is to characterize DILI healthcare utilization within a single, large hospital system, correlating demographics and culprit medications to LOS and admission costs. These parameters were then compared to expected benchmarks derived from the Vizient Clinical Database.

Materials and Methods

Study Design

This retrospective cohort study was conducted at JHS between October 2020 and January 2025. JHS is a large healthcare network comprised of 32 hospitals that provide inpatient and outpatient services to the Mid-Atlantic region. Data collected by JHS includes demographic information, diagnoses, laboratory data, referral services, medications, and insurance status.

Study Patients and Confirmation of DILI

Potential DILI patients at JHS 18 years or older were identified using ICD codes, a standardized code used to classify diseases, symptoms, injuries, and causes of death. Cases are assigned an ICD code during or after a patient encounter based on a provider’s judgement of patient diagnosis, symptoms, or reason for the visit. While no ICD code specifically indicating DILI exists, cases suggesting potential DILI include K710 (toxic liver disease with cholestasis), K711 (toxic liver disease with hepatic necrosis), K712 (toxic liver disease with acute hepatitis), K716 (toxic liver disease with hepatitis, not elsewhere classified), K719 (toxic liver disease, unspecified), K720 (acute and subacute hepatic failure), K729 (hepatic failure, unspecified), K759 (inflammatory liver disease, unspecified), and K762 (post-procedural hepatic failure). Relevant JHS sites included those utilizing Epic electronic health record (EHR). Cases included in this study were diagnosed at nine JHS sites.

All JHS patients with potential DILI had their records reviewed by a trained study contributor. For each patient, medical records, laboratory results, gastroenterologist/hepatologist consultation notes, and hospital discharge notes were examined. Due to the reliance on clinical suspicion and exclusion of other etiologies for DILI diagnosis, only cases explicitly noting DILI were included. All exclusion criteria are shown in Figure 1, including shock liver, graft versus host disease, and cases where a relevant identifying ICD code was observed outside of the time period for the study. Furthermore, cases in which there was high clinical suspicion for an etiology other than DILI were excluded. Included cases were then classified as cholestatic, hepatic, or mixed based on appropriate laboratory values per clinical practice guidelines.^12,13

Figure 1 Flow Diagram of DILI Cases Included in the Study. Abbreviation: ICD, International Classification of Diseases. Note: some cases met more than one exclusion criteria.

Vizient Database

Vizient Clinical Database is a platform that provides clinical benchmarks indicating hospital quality and financial performance. It provides data on patient outcomes compared with peer institutions. Based on patient data submitted to the database, Vizient calculates expected values for key healthcare system performance parameters, namely length of stay (LOS), in-hospital mortality, admission cost, and readmission rates. More specifically, expected values for cost are generated in the Vizient platform through proprietary models that aggregate pharmacy, service, and supply cost data from all contributing institutions and incorporate regional factors such as labor availability, patient population, and facility characteristics. Similarly, expected LOS is estimated from past similar encounters among contributing institutions. Through these models, the Vizient database provides healthcare utilization estimates for patient cases given a patient’s diagnosis and demographic characteristics. The present study subtracted observed values for LOS and cost from their respective expected values from the Vizient database and reported differences in these key healthcare utilization parameters for analysis.

Statistical Analysis

Incidence rates for DILI within JHS were calculated by dividing the number of confirmed DILI inpatient cases by the total number of inpatient cases at relevant JHS sites. Among all confirmed DILI cases, the differences between expected and observed cost and LOS were compared using paired t-tests. The normality of the distribution for cost and LOS differences was confirmed visually. Median differences between observed and expected values were reported when individual cases were grouped for further analysis. For LOS and cost, positive results indicate higher than expected utilization, and negative results represent lower than expected utilization. JMP Pro version 18.0.2 was used for all statistical analyses.

Ethical Considerations

This study was evaluated by Institutional Review Board (IRB) #153 at Thomas Jefferson University and was determined to be exempt from review (2024–3026). This was a retrospective study, and patient consent was not required by our IRB as patient interaction was not required; individual informed consent was waived because the study data was de-identified or originated from a registry. This study was conducted in accordance with the principles of the Declaration of Helsinki.

Results

Relevant patient characteristics and healthcare utilization parameters for the included DILI cases are detailed in Tables 1 and 2. The incidence of DILI at JHS was 9.19 per 100,000 hospitalized persons during the study period. In Table 3, a statistically significant increase in treatment cost was noted when comparing cases at JHS with benchmarks provided by Vizient. LOS also showed an increase at JHS versus Vizient benchmarks, although statistical significance was not reached. These results indicate high resource utilization at JHS compared to national trends. However, two cases were identified that required particularly high LOS and cost, indicating the need for caution in result interpretation.

Table 1 Characteristics of 48 Inpatient Members Confirmed to Have DILI at JHS During the Time Period

Table 2 Healthcare Utilization Parameters of 48 Inpatient Members Confirmed to Have DILI at JHS During the Time Period

Table 3 Mean Aggregated Healthcare Utilization Parameters of 48 Patients Confirmed to Have DILI at JHS During the Time Period, Compared to Vizient Benchmarks

Analysis of demographic trends in DILI revealed substantial healthcare utilization in key populations (Table 4). Females required higher than expected median LOS and cost, while males required lower than expected LOS and cost. A higher incidence of DILI was also noted in females. Treatment for those identifying as black involved the highest median difference in LOS and cost even though DILI was most common in those identifying as white. Collectively, these results indicate higher than expected healthcare utilization in historically marginalized populations.

Table 4 Median Aggregated Healthcare Utilization Parameters of 48 Patients Confirmed to Have DILI at JHS During the Time Period, By Demographic Variables

As noted in Table 5, individuals with Medicare had the longest median LOS difference, aligning with known trends of increased DILI severity in elderly populations.¹⁴ However, increased LOS did not translate to higher-than-expected median cost. The commercial/private payer and Medicaid groups showed comparable median LOS differences, although there was greater variation in LOS in the Medicaid group. The highest cost difference was observed in treating those with Medicaid.

Table 5 Median Aggregated Healthcare Utilization Parameters of 48 Patients Confirmed to Have DILI at JHS During the Time Period, By Payer Status

The highest number of DILI cases were due to antimicrobials, multiple agents, or agents not otherwise categorized (Table 6). DILI due to HDS was the least resource intensive to treat, although the greatest variation in cost was observed in the HDS group. Cases with unclear causative agents, whether the “multiple agent” or “other agent” categories, yielded the highest median LOS and cost difference.

Table 6 Median Aggregated Healthcare Utilization Parameters of 48 Patients Confirmed to Have DILI at JHS During the Time Period, By Causative Agent

Hepatic pattern DILI was the most commonly observed in the present study, followed by cholestatic and mixed pattern as indicated in Table 7. Both hepatocellular and mixed patterns exhibited a higher than expected LOS difference, while cholestatic was lower than expected. The mixed pattern yielded a median cost that was higher than expected, while both hepatic and cholestatic patterns had lower than expected cost.

Table 7 Median Aggregated Healthcare Utilization Parameters of 48 Patients Confirmed to Have DILI at JHS During the Time Period, by DILI Pattern

Discussion

In this study, we measured the economic burden of a relatively rare drug reaction to elucidate its impact on an academic, single hospital system in comparison to national trends. As shown in Table 3, we found an increased mean cost of admission for DILI compared with the mean expected cost from the national Vizient database. The length of stay, while higher than expected, was not statistically significant. This observed difference could be due to trainees ordering more advanced testing at academic medical centers in response to ambiguity of presentation. This result corresponds with a cross-sectional study in 2018 that found that major teaching hospitals ordered significantly more lab tests per day for pneumonia and cellulitis in comparison to non-teaching hospitals, even when adjusting for severity and demographics.¹⁴ A study in 1998 also found academic teaching hospitals were 63% more costly per inpatient case than non-teaching hospitals.¹⁵

Another hypothesis could be the attraction of more complex cases or administration of more expensive treatments, as seen in a study comparing costs between teaching and non-teaching hospitals for cancer treatment.¹⁶ When reviewing the results in the present study, two cases were significantly more expensive than the others. Both cases initially presented with transaminitis but had co-morbid conditions that predisposed them to develop severe infections, complicating their hospital course. This supports the theory that cases of DILI within JHS can be complicated and broad in clinical presentation, requiring high resource utilization.

Interestingly, when comparing cost and LOS among sexes, females were found to have a higher than expected admission cost and LOS than males (Table 4). This corroborates previous descriptions of sex differences in DILI, with females reportedly having an increased severity and mortality, and a 1.5–1.7 fold increased risk for developing an adverse drug reaction.^17–19 While cost is not a direct measurement of severity, it is expected to increase proportionally with case severity. Another demographic with higher than expected costs was black patients. Results showed that they had the highest median difference in LOS and cost despite the greater frequency of DILI in Caucasian patients. This supports studies that have shown a higher rate of hospitalization, liver transplantation, and mortality from DILI in black compared to white patients.²⁰

When categorizing based on payer status in Table 5, patients with Medicare and private insurance had lower than expected costs, whereas patients with Medicaid had higher than expected costs. Therefore, the relationship between private and public insurance is unclear but may point to insurance not being an accurate indicator of resource utilization.

When grouping admission costs and LOS by causative agents in Table 6, cases in the “multiple agents” and “other agents” categories were the most resource intensive. This is possibly due to diagnostic uncertainty in the absence of a clear causative agent, resulting in increased testing. DILI secondary to HDS was found to have the lowest associated cost and LOS. There has been a rising prevalence of usage of HDS with associated DILI.^21,22 The lower cost of HDS-associated DILI may be secondary to the type of patients who consume HDS, which tend to be health-conscious individuals, although one study found HDS-induced DILI tends to be more severe than other types of DILI.²³ Similarly, while antimicrobials were the most frequent causative agent implicated in DILI, they were also found to have low median cost and LOS difference. This may be due to the well-known risk of hepatotoxicity associated with antibiotics leading to prompt diagnosis. Lastly, DILI due to acetaminophen required shorter than expected median hospital stays, but higher than expected median costs. This again points to familiarity of acetaminophen toxicity, but the increase in cost may highlight a difference in management compared to national trends that necessitates more resources.

Some studies in the literature have correlated the pattern of injury with severity, with hepatocellular injury being associated with more severe DILI prognosis.¹⁹ In Table 7, we found that hepatocellular injury was the most common form of DILI, but it was not correlated with higher costs. In fact, a mixed injury pattern was the only one that resulted in higher costs. This may be due to delays in diagnosis, presumably because of ambiguity in interpreting liver function tests.

This study identifies several areas in which DILI has a higher than expected financial impact. Within the JHS hospital system, female patients, black patients, and DILI secondary to multiple possible agents were found to have a higher than expected cost compared to national trends. Further focus on these groups is necessary to identify the reasons for increased resource utilization. We hypothesize that the spectrum of disease, lack of familiarity of certain culprit medications, and complexity of the patient population with individualized risk factors may contribute to the unexpected cost differential. Studies have evaluated the efficacy of emerging biomarkers in combination with diagnostic scoring systems to enhance diagnostic and prognostic parameters, and it would be interesting to verify whether these new tools would reduce the cost of DILI.²⁴

A major limitation of this study is the small sample size, which is attributed to the uncommon prevalence of DILI in a single hospital system. Use of diagnostic codes to identify DILI has been shown to have reduced sensitivity in capturing DILI cases, likely leading to underreporting of cases within the timeframe.²⁵ Furthermore, DILI incidence at JHS was 9.39 cases per 100,000 hospitalized patients, which is lower than others reported in the literature.^1–3 Additional work is required to expand the cohort of DILI patients for more robust analysis. As mentioned above, we noted two cases that were more expensive than the median cost by nature of their complicated hospital course, which could contribute to the higher than expected costs in a small cohort. Lastly, while cost would presumably be higher in severe cases of DILI, we did not directly measure DILI severity for each case. Thus, further analysis is needed to determine whether cost is mostly dictated by severity, or if there are other individual or systemic factors that contribute to the financial impact of DILI.

Conclusion

The present study characterized DILI healthcare utilization trends at JHS, compared key parameters to benchmarks from Vizient Clinical Database, and correlated demographics, payer status, and causative agents with LOS and cost. At JHS, certain subgroups of patients with DILI required more resources than expected compared with Vizient benchmarks. Historically marginalized groups, including females and those identifying as black, required higher than expected resources to treat, although insurance status was not strongly associated with healthcare utilization. Cases involving multiple agents or agents not included in a DILI-causing category were the most resource expensive, likely due to additive drug effects and lack of familiarity with uncategorized therapeutics. Mixed pattern DILI required the highest healthcare utilization, whereas HDS cases were found to be relatively resource minimal. This study collectively suggests that clinical suspicion for DILI and swift determination of the offending agent are key to minimizing resources required for treatment. Social determinants of health likely contribute to healthcare utilization in DILI, although expansion of the cohort is required to further elucidate these trends.