Acute pancreatitis following treatment with protease inhibitors, which may be potential therapeutics for COVID-19: A real-world analysis of postmarketing surveillance data

Backgrounds: The potential therapeutic effects of protease inhibitors (PIs), such as lopinavir/ritonavir and darunavir, on COVID-19 are being tested in clinical trials. Although acute pancreatitis (AP) has been reported in patients treated with PIs, there have been few real-world studies comparing the occurrence and characteristics of AP after different PI regimens. Methods: Disproportionality analysis and Bayesian analysis were utilized for data mining of the Food and Drug Administration's Adverse Event Reporting System (FAERS) database for suspected adverse events involving AP after PI from January 2004 to December 2019. The times to onset and fatality rates of AP following different PI regimens were also compared. Results: Based on 33,832 reports related to PIs, 285 cases (0.84% of total adverse drug reactions, ADRs) were associated with AP; in these reports, the number of AP cases reported for the top �ve PIs was as follows: ritonavir/dasabuvir/ombitasvir/paritaprevir, 64 (22.46%); ritonavir, 54 (18.95%); atazanavir, 52 (18.25%); lopinavir/ritonavir, 48 (16.84%); and darunavir, 26 (9.12%). Twelve out of the 15 studied PIs, including lopinavir/ritonavir, darunavir and nel�navir, which are potential therapeutics for COVID-19, were associated with AP. Of all the reported adverse events involving AP related to PIs, 64.56% occurred in men, which was a much higher proportion than what was observed in women (28.42%). The median time to onset of AP was 103 (IQR: 26-408) days after the initiation of PI treatment. Patients treated with ritonavir/dasabuvir/ombitasvir/paritaprevir appeared to have an earlier onset of AP than those receiving atazanavir (31 [IQR: 17–68.25] days vs 187.5 [IQR: 80.5–556.5] days, p=0.0379) or ritonavir (31 [IQR: 17–68.25] days vs 177 [IQR: 56–539] days, p=0.0371). Compared with AP cases induced by all studied PIs, which had a fatality rate of 14.02%, AP cases associated with ritonavir (18.87%) and lopinavir/ritonavir (22.73%) appeared to be associated with a higher risk of death. Conclusions: Analysis of the FAERS data provides a more precise understanding of the occurrence and characteristics of AP after different PI regimens. Signals for AP associated with various PI regimens have been detected. The �ndings support continued surveillance, risk factor identi�cation, and comparative studies.


Introduction
As of May 24, 2020, coronavirus disease 2019 (COVID- 19), which is caused by a novel coronavirus (named SARS-CoV-2 by WHO) and was rst identi ed in Wuhan, China, has spread globally.The number of globally con rmed cases is more than 5204,000 as of the submission of this manuscript [1] .To date, there are no speci c treatments for COVID-19.Identifying effective and safe drugs to treat the virus as soon as possible is critical for the response to the COVID-19 outbreak.
The combination of lopinavir and ritonavir is widely used as a pharmacologically boosted protease inhibitor (PI) in highly active antiretroviral therapy (HAART) for human immunode ciency virus (HIV)-infected patients [2,3] .Due to its demonstrated e cacy in treating pneumonia associated with severe acute respiratory syndrome coronavirus (SARS-CoV) [4] and Middle East respiratory syndrome coronavirus (MERS-CoV) [5,6] in addition to its activity against SARS-CoV-2 in vitro, lopinavir/ritonavir has been applied to treat COVID-19, and some cases have been reported [7,8] .Triple combination therapy with ritonavir/lopinavir, ribavirin and interferon-alpha has been recommended as a drug treatment option by the guidelines for the diagnosis and treatment of COVID-19 (trial version 7), which was published by the National Health Commission of China.Nel navir, another PI that can strongly inhibit the replication of SARS-CoV [9] , may also be a candidate drug for COVID-19.
In the absence of a speci c treatment for COVID-19, several subsequent clinical trials (ChiCTR2000030187, ChiCTR2000029741, ChiCTR2000029603, ChiCTR2000029573, ChiCTR2000029548, ChiCTR2000029541, ChiCTR2000029539, ChiCTR2000029468, ChiCTR2000029387, and ChiCTR2000029308) have been quickly conducted in China to test the e cacy and safety of lopinavir/ritonavir and darunavir/cobicistat (another boosted PI) as monotherapies or combination therapies for the treatment of COVID-19-associated pneumonia in more than 10 hospitals [10] .Under this circumstance, not only the e cacy but also the safety of these PIs for the treatment of COVID-19 still need to be further con rmed.
As part of HAART, PIs have led to a signi cant decline in HIV-related morbidity and mortality [11] .However, PI-based treatments are, in general, associated with changes in the distribution of body fat and metabolic disorders, such as insulin resistance, hypercholesterolemia, and hypertriglyceridemia [12][13][14][15] .The latter is usually severe and di cult to control and may trigger episodes of AP [16] .PI-associated pancreatitis, some cases of which were fatal, has been observed [17,18] .The risk of severe pancreatitis induced by PI therapy has attracted a great deal of attention.Previous studies have evaluated whether the etiology of AP has changed since PI-based HAART therapy was rst introduced in 1996 [19][20][21][22] .However, past studies have reported con icting results regarding the impact of PIs on the development of pancreatitis [16] .Moreover, most evidence supporting the causal relationship between AP and PIs has come from case reports [23][24][25] and is insu cient to provide an overview of the risk of rare adverse events such as AP.
Pharmacovigilance studies regarding PI-induced AP events are still scarce.In particular, we still know little about the safety pro le of PI regimens in regard to pancreatitis in clinical practice.Characterizing PI-associated pancreatitis may provide important clues for patient safety during antiviral therapy.By analyzing adverse events extracted from the FDA's Adverse Event Reporting System (FAERS) database, we focused on evaluating and comparing the associations between various PI regimens and AP in a large real-world patient population.In addition, we investigated fatality rates and the times to onset of AP after the administration of various PI regimens.

Data source
We conducted a retrospective pharmacovigilance study using the FAERS database data from January 2004 to December 2019.The FAERS is a public voluntary spontaneous reporting database that provides information on adverse event and medication error reports submitted by health professionals, consumers, and manufacturers both domestically and abroad.
The ASCII data files of the FARES database contain demographic and administrative information (DEMO), adverse event (REAC), drug information (DRUG), outcomes (OUTC), report sources (RPSR), therapy information (THER) and using indications (INDI).
A total of 13,649,428 reports were extracted from the FAERS database.According to the FDA's recommendations, a deduplication procedure was performed to select the latest FDA_DT with the same CASEIDs and select the higher PRIMARYID when the CASEID and FDA_DT were the same, resulting in 1,145,704 reports.According to the PRIMARYID of the deduplicated DEMO data, the DRUG and REAC data were deduplicated as well.

Adverse event and drug identification
All 21 preferred terms (PTs) describing adverse events, such as "Pancreatitis relapsing" and "Pancreatitis acute", in the REAC files were investigated using the Medical Dictionary for Regulatory Activities(MedDRA) V23.0 and the Standardised MedDRA Queries (SMQ) term "acute pancreatitis" (SMQ: 20000022).
Drug names can be input arbitrarily in the FAERS database; therefore, MICROMEDEX® (Index Nominum) was utilized as a dictionary for the names of PIs (Table 1).
Both generic names and brand names were used as keywords for FAERS database retrieval.

Data mining
The disproportionality analysis (the reporting odds ratio, ROR and the proportional reporting ratio, PRR) and the Bayesian analysis (the Bayesian confidence propagation neural network, BCPNN and the multi-item gamma Poisson shrinker, MGPS) algorithms were used to identify statistical associations between a drug and an adverse event (Table 2) [26-34] .In this study, adverse events were extracted when at least 1 of 4 indices met the criteria.IC025=e ln(IC)-1.96(1/a+1/b+1/c+1/d)^0. 5 MGPS EBGM=a(a+b+c+d)/((a+c)(a+b)) EBGM05>2, N>0 EBGM05=e ln(EBGM)-1.64(1/a+1/b+1/c+1/d)^0.The associations between AP and different PI regimens were compared.The time to onset of AP was evaluated for each PI regimen.It was defined as the interval between the EVENT_DT (adverse event onset date) and the START_DT (start date of PI use).However, reports with input error (an EVENT_DT earlier than the START_DT) or inaccurate date entry were excluded.In addition, reports with fatal events attributed to drug toxicity were counted, and the fatality rate was calculated as the number of fatal events divided by the total number of events of AP associated with each PI regimen.

Statistical analysis
Descriptive analysis was used to summarize the clinical characteristics of the patients with PI-associated AP collected from the FAERS database.The times to onset of PI-associated AP between different PI regimens were compared using nonparametric tests (the Mann-Whitney test for dichotomous variables and the Kruskal-Wallis test when there were more than two subgroups of respondents) because the data were not normally distributed.Fatality rates were compared among different PI regimens using Pearson's chi-square test.The statistical significance was determined at p < 0.05 with 95% confidence intervals.Data mining and all statistical analyses were performed with SAS, version 9.4 (SAS Institute Inc., Cary, NC, USA).

Descriptive analysis
33,832 adverse events related to PIs and 51,883 reports related to AP were recorded in the FAERS database between January 2004 and December 2019.Among them, PIs were set as the suspected drugs that caused AP in 285 reports, and the clinical characteristics of these patients are presented in Table 3.The highest number of AP reports was associated with ritonavir/dasabuvir/ombitasvir/paritaprevir (n=64, 22.46%), followed by ritonavir (n=54, 18.95%), atazanavir (n=52, 18.25%), lopinavir/ritonavir (n=48, 16.84%), and darunavir (n=26, 9.12%).AP adverse events were most reported in HIV-infected patients (64.77%).Of these events, 64.56% occurred in men, and 28.42% occurred in women; the gender was unknown in 7.02% of the events.All but two events involved adult patients (aged > 18 years), and the average age was 49.67 years.Most of the cases were from North America (40.35%) and Europe (30.88%) and were mainly submitted by health care professionals (71.93%).

analysis and Bayesian analysis
No cooccurrence of use and AP was reported, and only 1 and 2 cooccurrences of PI use and AP were reported for darunavir/cobicistat/emtricitabine/tenofovir and darunavir/cobicistat, respectively.Overall, according to the criteria for the 4 algorithms, 12 out of 15 PIs (all except the ones mentioned above) were suggested to be associated with AP.Lopinavir/ritonavir, ritonavir, nelfinavir, and darunavir, which are potential therapeutics for COVID-19, appeared to be associated with AP.The results are listed in Table 4.Among all PI therapies, indinavir was notably associated with AP because it had the highest ROR, PRR, IC, and EBGM.

Discussion
Since the outbreak of COVID-19, there have been more than 5204,000 con rmed cases globally, and 337,687 people have lost their lives [1] .Thousands more are ghting for their lives in hospitals.The World Health Organization (WHO) has characterized COVID-19 as a pandemic.All countries activated and scaled up their emergency response mechanisms, and a great deal of effort has been made to identify effective drugs against the virus.Some PIs, such as lopinavir/ritonavir and darunavir/cobicistat, have been tested in clinical trials in China.This study focused on the occurrence, timing, and prognosis of AP following the use of various PI regimens in real-world practice based on the FAERS pharmacovigilance database.First using disproportionality analysis and Bayesian analysis as a rapid and effective signal detection method to perform the largest-to-date postmarketing surveillance of these PIs, we attempted to provide valuable and timely signals for clinical evaluation to minimize the potential harm induced by AP in the treatment of COVID-19.The study showed that 12 out of the 15 studied PIs, including lopinavir/ritonavir, darunavir, and nel navir, which are potential therapeutics for COVID-19, were associated with APrelated adverse events.
Despite the potential bene ts it provides to COVID-19 patients, the use of PIs may be accompanied by serious side effects, including AP.Two case reports [23,24] have suggested a causal mechanism between PI-induced hypertriglyceridemia and AP, while a third case report of PI-associated acute pancreatitis was not associated with a rise in serum triglycerides [25] .It remains unclear whether PIs are associated with pancreatitis.Hence, it is important to recognize the associations between particular PI regimens and AP as well as the clinical features and develop an awareness of this adverse event among practitioners prescribing PI for the treatment of COVID-19 and other professionals, such as pharmacists and nurses.However, the assessment of PI-associated AP is quite challenging because of its low incidence and overlooked manifestations [21,22] .Due to strict study entry criteria, relatively small sample sizes, and a nite scope and time frame, it is di cult to achieve this goal through experimental studies, including mandatory clinical trials, alone.Performing postmarketing surveillance is an e cient way to discover rare but potentially severe adverse reactions and help us acquire vital basis for prevention.Therefore, our study provides a pro le of PI-associated AP through data mining of 33,832 adverse events related to PIs documented in the FAERS database.
Some prior studies that included patients receiving HAART found an association between female gender and AP and hypothesized that a smaller body weight might increase the toxicity of nucleotide reverse-transcriptase inhibitors (NRTIs) [35,36] .However, Manfredi et al. found no association between gender and the risk of AP related to PI-based HAART exposure [21] .In our study, based on reports from the FAERS database, we found that PI-associated AP seemed to predominately affect men rather than women (64.56% vs. 28.42%).Most of these collected reports were from North America (40.35%) and Europe (30.88%), where women make up a smaller proportion of the HIV patient population [37] .Therefore, we could not conclude that men are more likely than women to suffer from AP following PI regimens.Further research is required to reevaluate the relationship between gender and PI-associated AP.
Guo et al. found that advanced age is a risk factor for acute pancreatitis after a retrospective cohort study of 4,972 patients with HIV infection, in which they found 159 cases of AP [38] .Conversely, we did not observe that PI-induced AP predominately affected elderly patients (11.93% of patients ≥65 years vs. 53.68% of patients < 65 years).On account of a lack of information in the FAERS database, it was di cult to control confounding factors such as age in this study.Further research is needed to explore the association between advanced age and the incidence of PI-related AP.
Previous studies have reported con icting results regarding the impact of PIs on the development of pancreatitis [16] .Most evidence supporting the causal relationship of AP to PIs has come from case reports [23][24][25] and is insu cient to provide an overview of the risk of rare adverse events such as AP.Riedel et al.
found that neither PI nor non-nucleoside reverse-transcriptase inhibitors (NNRTIs) are associated with an increased rate of AP through a ten-year cohort study [39] .Conversely, based on this pharmacovigilance analysis, 12 out of the 15 studied PIs were found to be associated with AP-related adverse events, which is similar to what was reported in some prior studies [21,22] .Hypertriglyceridemia has been demonstrated to cause pancreatitis [40,41] .According to previous studies, all available PIs are associated with signi cant increases in plasma triglyceride concentrations [42,43] , but hypertriglyceridemia is more frequently observed following the use of ritonavir or lopinavir/ritonavir combination therapy than following other PI-based combinations [44,45] .Due to their potential clinicopathological consequences, ritonavir and lopinavir/ritonavir regimens may lead to an increased risk of hyperlipidemic pancreatitis [42,46] .Surprisingly, we observed that indinavir, not ritonavir, seemed to have the strongest association with AP among all PI regimens, including ritonavir and lopinavir/ritonavir.This may be due to the relatively limited cases of AP-associated indinavir reported (8 reports).In our study, the 64.77% of reported cases are involving HIV-infected patients.As we know, antiretroviral treatment is very complex and dangerous due to concomitant medications and their toxic effects, including drug-induced AP.There are plenty of factors which can affect the pancreas, such as a previous history of AP, hepatobiliary diseases, alcohol abuse, low CD4 counts, and opportunistic infection prophylaxis, in addition to direct lesions caused by HIV [16] .All these factors make the evaluation of the association between PI and AP more di cult.
Another nding was that the median time to the onset of AP was 103 (IQR: 26-408) days after the initiation of PI treatment, which was similar to the onset time observed in some previous case reports [23][24][25] .Ritonavir/dasabuvir/ombitasvir/paritaprevir appeared to have the earliest onset of AP (31 [IQR: 17-68.25]days) among all studied PI regimens.For ritonavir and lopinavir/ritonavir, the median times to onset were 177 (IQR: 56-539) days and 91 (IQR: 22-352.5)days, respectively.It was suggested that ritonavir-based PI regimens may induce AP earlier than other PI regimens.Although the onset of AP after the use of PI seemed to be long according to some reports collected in this study, AP could occur as soon as after the rst several doses of the ritonavir-based PI regimens mentioned above and atazanavir.Notably, triple combination therapy with ritonavir/lopinavir, ribavirin and interferon-alpha has been recommended as a drug treatment option by the guidelines, and it might be predominantly used to treat COVID-19 patients.However, both interferon monotherapy and interferon/ribavirin combination therapy have been reported in the literature to be causally associated with AP [47][48] .As a consequence of the multifactorial effects, triple combination therapy may lead to an increased risk of AP.Patients receiving PI regimens or the recommended triple combination therapy for the treatment of COVID-19 should be closely observed.Drug-induced AP should be considered, if patients present with clinical signs and symptoms of pancreatitis, including abdominal pain, nausea, vomiting, and conjunction with abnormal levels of serum lipase or/and amylase.
We also assessed and compared the fatality rates of AP associated with PIs to investigate differences in the severity of AP associated with various PIs.It was observed that AP was generally associated with poor outcomes, exhibiting a fatality rate of 14.02%, which was more than two times that found by Riedelet al.
in a cohort study that followed 5,970 HIV-infected patients [39] .In that study, ve patients died of pancreatitis during their hospitalization, yielding an in-hospital mortality rate of 5.9% for AP.As there is no consensus de nition of acute pancreatitis, the different inclusion criteria used in the two studies may have resulted in different speci cities and sensitivities of case inclusion.This may have caused the difference between the fatality rates in the two studies.Furthermore, we found that among all 271 reported PI-induced AP cases, compared with the other PIs, ritonavir (18.87%) and lopinavir/ritonavir (22.73%) appeared to be associated with a higher risk of death.However, based on the collected data, there was no signi cant difference in fatality rates across different PI regimens (Pearson's chi-square test for overall comparison, p=0.730).Fatal events occurred in 1 (25.00%) of 4 patients treated with saquinavir and 1 (20.00%) of 5 patients treated with nel navir; however, there were not enough cases reported to draw a conclusion, and continued surveillance is needed.
Based on spontaneous reporting systems (SRSs), disproportionality analysis and Bayesian analysis allow for signal detection rapidly and generate hypothesis about associations between AP and various PI regimens.However, it should be noted that in addition to many advantages, data mining techniques used in this study still have several limitations.First, affected by the inherent limitation of SRSs, sources of AP reports are non-homogeneous.Incomplete reporting, underreporting, false reporting, and inaccuracy might result in reporting bias and even misleading [49] .Second, owing to a lack of insu cient information in the FAERS database, the possibility of veri cation of the clinical ndings justifying the reported AP is quite limited, and it is also scarcely possible to control such confounding factors as indications, pre-existing pancreatic diseases, comorbidities or other factors which might have an impact on AP occurrence [50] .Third, as reports extracted from the FAERS database do not re ect the total number of adverse reactions involving AP, the number of reports for a particular PI may be in uenced by the extent of use of the product, which may be affected by some factors such as enterprise publicity and product price [49,50] .It cannot be used to quantify the incidence of PI-associated AP.Data mining techniques only provide a pro le of PI-associated AP through signal detection.It is generally insu cient to prove the causal relationship, which needs to be replicated ideally by prospective studies [51] .

Conclusion
Based on real-world practice data, this pharmacovigilance study identi ed associations between AP and various PI regimens, including lopinavir/ritonavir, darunavir and nel navir, which are potential therapeutics for COVID-19.Indinavir appeared to have a stronger association with AP than other PIs.Moreover, AP associated with ritonavir boosted regimens had an earlier onset and occurred as soon as after the rst several doses in some cases.Further pharmacoepidemiological studies should continue to evaluate the hypotheses generated by this study.Our ndings provide a foundation for continued surveillance and investigation into this matter.
Not only patients receiving long-term PI treatment, but also those receiving PI regimens for the short -term treatment of COVID-19 should be closely observed.

5 Abbreviations
Abbreviations: a: number of reports containing both the suspected drug and the suspected adverse drug reaction; b: number of reports containing the suspected adverse drug reaction with other medications (except the drug of interest);.c: number of reports containing the suspected drug with other adverse drug reactions (except the event of interest); d: number of reports containing other medications and other adverse drug reactions; ROR: reporting odds ratio; CI: confidence interval; N: number of cooccurrences of PI use and AP; PRR: proportional reporting ratio; χ2: chi-squared; BCPNN: Bayesian confidence propagation neural network; IC: information component; IC025: the lower limit of the 95% two-sided CI of the IC; MGPS: multi-item gamma Poisson shrinker; EBGM: empirical Bayesian geometric mean; EBGM05: the lower 90% one-sided CI of EBGM Figures

Table 1 :
Summary of FDA-approved protease inhibitors

Table 2 :
Summary of Major Algorithms Used for Signal Detection

Table 3 :
Clinical characteristics of patients with protease inhibitor-associated acute pancreatitis collected from the FAERS database

Table 4 :
Disproportionality analysis and the Bayesian analysis of associations of protease inhibitor regimens with acute pancreatitis Time to onset of PI-associated AP The overall median time to event onset of PI-related AP was 103 (interquartile range [IQR]: 26-408) days.For ritonavir, lopinavir/ritonavir, darunavir and nelfinavir, the median times to onset were 177 (IQR: 56-539) days, 91 (IQR: 22-352.5)days,311 (IQR: 235-481.5)daysand 918 (IQR: 469-2132) days, respectively.The times to onset associated with each PI regimen are shown in Fig. 1.To evaluate the prognosis of AP after PI use, we assessed fatality rates due to AP-related adverse events following various PI regimens, and the results are shown in Abbreviations: PI: protease inhibitor; N: number of reports of PI-associated acute pancreatitis; ROR: reporting odds ratio; CI: confidence interval; PRR: proportional reporting ratio; χ 2 : chi-squared; IC: information component; EBGM: empirical Bayes geometric mean3.3.

Table 5 .
Fatalities occurred in 38 (14.02%) of 271 reported PI-induced AP cases.There was no significant difference in fatality rates across different PI regimens (Pearson's chi-square test for overall comparisons, p=0.730).

Table 5 :
rates due to acute pancreatitis adverse events following various protease inhibitor regimens