Assessing Drug, Herbal, and Dietary Supplement Hepatotoxicity
Assessing Drug, Herbal, and Dietary Supplement Hepatotoxicity
The DILIN study has been previously described in detail. Patients suspected of having liver injury owing to medications or HDS products were enrolled within 24 weeks of injury onset and then followed up prospectively for 6–24 months depending on the pace and completeness of DILI resolution. Because the enrolment window was 24 weeks, cases enrolled at varying time points in their DILI event. Enrolment criteria were serum alanine aminotransferase (ALT) or aspartate aminotransferase (AST) levels > 5 times the upper limit of the normal (ULN) (or pretreatment baseline if abnormal) on two consecutive occasions, or alkaline phosphatase (AP) levels > twice the ULN (or pretreatment baseline if abnormal) on two consecutive occasions, or total serum bilirubin > 2.5 mg/dl, or international normalized ratio (INR) >1.5 with any elevation in serum ALT, AST or AP. Cases meeting laboratory enrolment criteria are enrolled at the discretion of the investigators at each site and based on their clinical opinion. Case enrolment is not restricted by time of onset because some medications are well known to have very long latencies of even years (e.g. nitrofurantoin). We did not restrict enrolment to those with full follow-up data showing resolution of injury (i.e. dechallenge) because such restriction would hinder the prospective nature of DILIN. We wanted to capture this dechallenge data while under study protocol. Injury pattern was categorized as cholestatic (R < 2), mixed (R 2–5) or hepatocellular (R > 5) where R = (ALT/ULN) ÷ (AP/ULN). Severity level was based on INR, bilirubin, signs of liver failure, need for hospitalization and fatal or transplant outcome as previously described. Exclusion criteria included acetaminophen hepatotoxicity, prior liver or bone marrow transplant, alcohol-related liver disease, autoimmune hepatitis or genetic liver disease. Patients with compensated chronic hepatitis B, C or with nonalcoholic fatty liver disease were eligible and enrolled at the discretion of the site investigators. For patients with such background liver disease, reviewers used baseline liver enzyme levels, viral serologies and viral nucleic acid tests to judge the presence of DILI vs. exacerbation of underlying liver disease.
At baseline visit, a detailed history was obtained, and clinical, laboratory and imaging results were extracted from records. As reported previously, a DILIN protocol battery of tests to exclude other causes of liver injury was obtained at enrolment if not extractable from chart records. Serum, plasma, urine and DNA specimens were sent to a central repository for future studies. Subjects were followed up for at least 6 months and those with persistent liver abnormalities or signs of chronic liver injury were followed up through 24 months.
Ninety (90%) of the cases had complete documentation of data for all 21 parameters adapted from Agarwal et al. as essential data for DILI cases. (Appendix figure) Nine (9%) were missing complete data for one parameter each (four cases – incomplete viral serologies, four cases – no documented hepatic imaging, one case – incomplete 'washout' of liver biochemistries); one case (1%) was missing complete data for two parameters, viral serologies and autoimmune markers. This last case was considered unlikely to be DILI in large part owing to lack of these data. Overall, the 90 cases with complete data were considered more likely to be DILI than the 10 cases with incomplete data (median DILIN scores of 2 vs. 3, respectively, P = 0.04).
(Enlarge Image)
Appendix Figure.
Completeness of data at adjudication for parameters adapted from Agarwal et al. (7).
*Laboratory values including liver biochemistries, INR, cell count at onset and enrollment.
**Complete viral serologies for acute hepatitis A, B and C.
^Autoimmune serologies (ANA, ASMA).
DILIN causality assessment has been described in detail. The process is the same for medication and HDS hepatotoxicity, as there are little data to suggest that a different causality process is necessary or valid. Each case was adjudicated independently by three hepatologists including the site investigator who enrolled the case. Assessment is based on retrospective review of the case history, laboratory data and prospective follow-up study visits. Each reviewer assigns a causality score corresponding to percentages of DILI likelihood in which 1 = definite (>95% likelihood), 2 = very likely (75–95%), 3 = probable (50–74%), 4 = possible (25–49%) and 5 = unlikely (<25%). Disagreements were identified after the three reviewers submitted their scores. Consensus scores were achieved by electronic mail or conference call discussions. Those cases in which agreement could not be reached by the three reviewers were then voted upon by one member from each DILIN site during monthly conference calls. Final score was assigned by majority vote. For cases involving >1 agent, an overall case score and separate individual scores were determined. For example, the overall case score might be 1 (definite DILI) with one agent scoring a 2 (very likely causal) and the other scoring a 4 (only possibly).
From inception in 2004 to April 16, 2009, adjudication was based on results obtained shortly after enrolment, thus simulating the clinician's task to assess at time of presentation. However, when enrolment occurred within days of onset, data on resolution of injury were sparse. After April 16, 2009, the protocol was changed, so that adjudication was done 6 months after enrolment and follow-up data could be included into the assessment.
The DILIN Data Coordinating Center chose 100 cases by computer-driven random assignment from the Prospective registry. Two cases involving the interval development of new diagnostic information regarding hepatitis E testing were included, but the HEV data were specifically excluded for reassessment. Chosen cases were stratified 1:1 across April 16, 2009. Group A included 49 cases enrolled before April 16, 2009, and Group B, 51 cases enrolled afterwards. Group A cases did not have 6-month data for the initial assessment, but these data were available for the reassessment. Group B cases had 6-month data for both the initial and reassessments. We stratified across these two periods to examine whether reliability is influenced by using follow-up data. For reassessments, 92 cases had three new reviewers. Because of an administrative error, eight cases had one previous reviewer and two new reviewers. No cases were reassessed by the site investigator who enrolled the case. The rationale for excluding the enrolling site investigator from reassessments was to minimize recall bias as the enrolling investigator is often the hepatologist who continues to care for the patient. At least 4 months had to elapse before a case could be selected for reassessment. The process of reassessment was otherwise the same.
Standard descriptive statistics were used to characterize this cohort of 100 patients and compared to the remaining 983 cases not selected. The original and re-adjudication scores were compared for the 100 cases using weighted kappa statistics owing to ordinal nature of the categories. Because some cases involved >1 drug or HDS product, the reliability at the individual agent level was also examined. Weighted kappa statistics were also determined for Group A and Group B cases separately. Because the two categories of highest likelihood (>95% likelihood and 75–95%) were similar in clinical and research relevance, reliability was assessed collapsing these two categories into one, thus creating quartiles of per cent likelihood. Score differences were categorized as 1, 2 or >2 scores apart and direction of changes were tallied. Cases with scores crossing the 50% likelihood line (1, 2 or 3 vs. 4 or 5) were re-examined in detail to assess factors that might have led to such uncertainty in DILI diagnosis.
Methods
DILIN Prospective Cohort
The DILIN study has been previously described in detail. Patients suspected of having liver injury owing to medications or HDS products were enrolled within 24 weeks of injury onset and then followed up prospectively for 6–24 months depending on the pace and completeness of DILI resolution. Because the enrolment window was 24 weeks, cases enrolled at varying time points in their DILI event. Enrolment criteria were serum alanine aminotransferase (ALT) or aspartate aminotransferase (AST) levels > 5 times the upper limit of the normal (ULN) (or pretreatment baseline if abnormal) on two consecutive occasions, or alkaline phosphatase (AP) levels > twice the ULN (or pretreatment baseline if abnormal) on two consecutive occasions, or total serum bilirubin > 2.5 mg/dl, or international normalized ratio (INR) >1.5 with any elevation in serum ALT, AST or AP. Cases meeting laboratory enrolment criteria are enrolled at the discretion of the investigators at each site and based on their clinical opinion. Case enrolment is not restricted by time of onset because some medications are well known to have very long latencies of even years (e.g. nitrofurantoin). We did not restrict enrolment to those with full follow-up data showing resolution of injury (i.e. dechallenge) because such restriction would hinder the prospective nature of DILIN. We wanted to capture this dechallenge data while under study protocol. Injury pattern was categorized as cholestatic (R < 2), mixed (R 2–5) or hepatocellular (R > 5) where R = (ALT/ULN) ÷ (AP/ULN). Severity level was based on INR, bilirubin, signs of liver failure, need for hospitalization and fatal or transplant outcome as previously described. Exclusion criteria included acetaminophen hepatotoxicity, prior liver or bone marrow transplant, alcohol-related liver disease, autoimmune hepatitis or genetic liver disease. Patients with compensated chronic hepatitis B, C or with nonalcoholic fatty liver disease were eligible and enrolled at the discretion of the site investigators. For patients with such background liver disease, reviewers used baseline liver enzyme levels, viral serologies and viral nucleic acid tests to judge the presence of DILI vs. exacerbation of underlying liver disease.
At baseline visit, a detailed history was obtained, and clinical, laboratory and imaging results were extracted from records. As reported previously, a DILIN protocol battery of tests to exclude other causes of liver injury was obtained at enrolment if not extractable from chart records. Serum, plasma, urine and DNA specimens were sent to a central repository for future studies. Subjects were followed up for at least 6 months and those with persistent liver abnormalities or signs of chronic liver injury were followed up through 24 months.
Ninety (90%) of the cases had complete documentation of data for all 21 parameters adapted from Agarwal et al. as essential data for DILI cases. (Appendix figure) Nine (9%) were missing complete data for one parameter each (four cases – incomplete viral serologies, four cases – no documented hepatic imaging, one case – incomplete 'washout' of liver biochemistries); one case (1%) was missing complete data for two parameters, viral serologies and autoimmune markers. This last case was considered unlikely to be DILI in large part owing to lack of these data. Overall, the 90 cases with complete data were considered more likely to be DILI than the 10 cases with incomplete data (median DILIN scores of 2 vs. 3, respectively, P = 0.04).
(Enlarge Image)
Appendix Figure.
Completeness of data at adjudication for parameters adapted from Agarwal et al. (7).
*Laboratory values including liver biochemistries, INR, cell count at onset and enrollment.
**Complete viral serologies for acute hepatitis A, B and C.
^Autoimmune serologies (ANA, ASMA).
DILIN Expert Opinion Process
DILIN causality assessment has been described in detail. The process is the same for medication and HDS hepatotoxicity, as there are little data to suggest that a different causality process is necessary or valid. Each case was adjudicated independently by three hepatologists including the site investigator who enrolled the case. Assessment is based on retrospective review of the case history, laboratory data and prospective follow-up study visits. Each reviewer assigns a causality score corresponding to percentages of DILI likelihood in which 1 = definite (>95% likelihood), 2 = very likely (75–95%), 3 = probable (50–74%), 4 = possible (25–49%) and 5 = unlikely (<25%). Disagreements were identified after the three reviewers submitted their scores. Consensus scores were achieved by electronic mail or conference call discussions. Those cases in which agreement could not be reached by the three reviewers were then voted upon by one member from each DILIN site during monthly conference calls. Final score was assigned by majority vote. For cases involving >1 agent, an overall case score and separate individual scores were determined. For example, the overall case score might be 1 (definite DILI) with one agent scoring a 2 (very likely causal) and the other scoring a 4 (only possibly).
From inception in 2004 to April 16, 2009, adjudication was based on results obtained shortly after enrolment, thus simulating the clinician's task to assess at time of presentation. However, when enrolment occurred within days of onset, data on resolution of injury were sparse. After April 16, 2009, the protocol was changed, so that adjudication was done 6 months after enrolment and follow-up data could be included into the assessment.
Reliability Cohort and Reassessment
The DILIN Data Coordinating Center chose 100 cases by computer-driven random assignment from the Prospective registry. Two cases involving the interval development of new diagnostic information regarding hepatitis E testing were included, but the HEV data were specifically excluded for reassessment. Chosen cases were stratified 1:1 across April 16, 2009. Group A included 49 cases enrolled before April 16, 2009, and Group B, 51 cases enrolled afterwards. Group A cases did not have 6-month data for the initial assessment, but these data were available for the reassessment. Group B cases had 6-month data for both the initial and reassessments. We stratified across these two periods to examine whether reliability is influenced by using follow-up data. For reassessments, 92 cases had three new reviewers. Because of an administrative error, eight cases had one previous reviewer and two new reviewers. No cases were reassessed by the site investigator who enrolled the case. The rationale for excluding the enrolling site investigator from reassessments was to minimize recall bias as the enrolling investigator is often the hepatologist who continues to care for the patient. At least 4 months had to elapse before a case could be selected for reassessment. The process of reassessment was otherwise the same.
Analysis
Standard descriptive statistics were used to characterize this cohort of 100 patients and compared to the remaining 983 cases not selected. The original and re-adjudication scores were compared for the 100 cases using weighted kappa statistics owing to ordinal nature of the categories. Because some cases involved >1 drug or HDS product, the reliability at the individual agent level was also examined. Weighted kappa statistics were also determined for Group A and Group B cases separately. Because the two categories of highest likelihood (>95% likelihood and 75–95%) were similar in clinical and research relevance, reliability was assessed collapsing these two categories into one, thus creating quartiles of per cent likelihood. Score differences were categorized as 1, 2 or >2 scores apart and direction of changes were tallied. Cases with scores crossing the 50% likelihood line (1, 2 or 3 vs. 4 or 5) were re-examined in detail to assess factors that might have led to such uncertainty in DILI diagnosis.
Source...