Predicting hepatotoxicity associated with low-dose methotrexate using machine learning

doi:10.21203/rs.3.rs-1753066/v2

Download PDF

Research Article

Predicting hepatotoxicity associated with low-dose methotrexate using machine learning

https://doi.org/10.21203/rs.3.rs-1753066/v2

This work is licensed under a CC BY 4.0 License

Version 2

posted

You are reading this latest preprint version

Purpose

An accurate prediction of the hepatotoxicity associated with low-dose methotrexate can provide evidence for a reasonable treatment choice. This study aimed to develop a machine learning-based prediction model to predict hepatotoxicity associated with low-dose methotrexate and explore the associated risk factors.

Methods

Eligible patients with immune system disorders, who received low-dose methotrexate at West China Hospital between January 1, 2018, and December 31, 2019, were enrolled. A retrospective review of the included patients was conducted. Risk factors were selected from multiple patient characteristics, including demographics, admissions, and treatments. Eight algorithms, including eXtreme Gradient Boosting (XGBoost), AdaBoost, CatBoost, Gradient Boosting Decision Tree (GBDT), Light Gradient Boosting Machine (LightGBM), Tree-based Pipeline Optimization Tool (TPOT), Random Forest (RF), and Artificial Neural Network (ANN), were used to establish the prediction model.

Results

A total of 782 patients were included, and hepatotoxicity was detected in 35.68% (279/782) of the patients. The Random Forest model with the best predictive capacity was chosen to establish the prediction model (receiver operating characteristic curve 0.97, accuracy 64.33%, precision 50.00%, recall 32.14%, and F1 39.13%). Among the 15 risk factors, the highest score was a body mass index of 0.237, followed by age (0.198), the number of drugs (0.151), and the number of comorbidities (0.144). These factors demonstrated their importance in predicting hepatotoxicity associated with low-dose methotrexate.

Conclusion

Using machine learning, this novel study established a predictive model for low-dose methotrexate-related hepatotoxicity. The model can improve medication safety in patients taking methotrexate in clinical practice.

Low-dose methotrexate

Hepatotoxicity

Machine learning

Prediction model

Risk factor

Methotrexate (MTX), a folic acid antagonist that inhibits dihydrofolate reductase in the S-phase cell cycle, was first developed as an anticancer treatment in the 1940s [1, 2]. Since the 1950s, MTX has been prescribed as an immunosuppressant for treating immune system disorders, including rheumatoid arthritis, psoriasis, psoriatic arthritis, and inflammatory bowel diseases [3]. The overall prevalence of rheumatoid arthritis is 0.24–1.1% [4–6]. Similarly, a population-based study in the United States found that psoriasis rates increased significantly from 50.8 cases per 100,000 (1970 to 1974) to 100.5 cases per 100,000 (1995 to 1999) [7]. A worldwide review showed that the prevalence of psoriasis ranged from 0.5 to 11.4% in adults [8]. MTX is now recommended as a first- or second-line treatment for many immune system diseases [9–15]. Although several biological agents have emerged for immune system diseases in the last two decades, such as adalimumab [16], infliximab [17], canakinumab [18], ustekinumab [19], and secukinumab [19], MTX is still widely used due to its efficacy, low cost, and ease of administration. MTX can be administered orally or subcutaneously as a weekly treatment regimen [11].

Due to lower doses of MTX, life-threatening adverse drug effects (ADE) are rarely observed in MTX treatment for immune system diseases. However, severe ADEs can still occur, especially hepatotoxicity [20]. Abnormal serum levels of alanine aminotransferase (ALT) or aspartate aminotransferase (AST) occurred in 23.47% (315/1,342) of patients treated for MTX-treated rheumatoid arthritis [21]. Liver enzyme abnormalities are the leading cause of dose modification or discontinuation of MTX [22]. Furthermore, a systematic review indicated that 33% of patients with psoriasis who received low-dose MTX had liver disease progression, such as liver fibrosis [23]. Therefore, it is crucial to clarify the incidence of liver ADE in MTX-treated patients. Risk factors can affect MTX therapies. These risk factors include alcohol use, history of liver disease, obesity, type 2 diabetes, history of significant exposure to hepatotoxic drugs or chemicals, lack of folate supplementation, and hyperlipidemia [24, 25]. However, current research lacks an assessment of the impact of these risk factors. Furthermore, there is a lack of research exploring unknown risk factors and establishing predictive models for hepatotoxicity associated with low-dose MTX.

Machine learning is one of the fastest-growing technical fields [26] and has been widely used in medical fields, such as medical diagnosis and prediction of disease risks [27–29]. Machine learning can promote data-driven estimation when selecting multiple variables and processing complex nonlinear relationships among multidimensional variables [30]. Therefore, machine learning can increase the precision of prediction models, especially for analyzing large datasets with many variables [31, 32]. The study aimed to compare eight machine learning methods to identify the most optimal model to predict hepatotoxicity and risk factors associated with low-dose MTX.

Study setting and the study population

This retrospective study was conducted at the West China Hospital of Sichuan University, a large tertiary teaching hospital in China. This hospital uses an electronic medical record (EMR) and bar code systems to document medication administrations. The study inclusion criteria were 1) patients with immune system diseases and 2) treated with low-dose MTX (≤ 30 mg per week) [33] during hospital stays at the West China Hospital of Sichuan University. Patients who were treated with other doses of MTX were excluded. The study period was from January 1, 2018, to December 31, 2019.

Data extraction

A two-stage review process for medical records was conducted to identify the presence of hepatotoxicity. In the first stage, two trained clinical pharmacists (Hu and Wang) independently reviewed each medical record for the presence of hepatotoxicity. The following sections of the charts were reviewed: basic patient information, diagnostic and progress notes, medication charts, laboratory data, surgical records, nursing flow sheets, and admission and discharge documents. In the second stage, a physician reviewed all medical records identified in the first stage to determine the presence of hepatotoxicity. Disagreements were resolved through a team discussion. We also reviewed the duplicate records to identify hepatotoxicity. For patients with hepatotoxicity, the medical records with hepatotoxicity were included in the analysis, otherwise the first medical records were used for analysis. Because the clinical interventions usually preceded patients reaching clinical diagnostic criteria of drug-induced liver injury [34], hepatotoxicity was defined as elevated liver enzymes > 1.25 of the upper limits of normal (ULN) ,liver failure, fibrosis, cirrhosis, or death due to drug-induced liver injury. The hospital standard cut-off values for ALT are 40 IU/L for women and 50 IU/L for men. AST values are 35 IU/L for women and 40 IU/L for men. Alkaline phosphatase (ALP) values are 135 IU/L for women and 160 IU/L for men. Gamma-glutamyl transpeptidase (GGT) values are 35 IU/L for women and 60 IU/L for men. Data collection was carried out from January to April 2022.

Based on data from included patients’ records, risk factors were screened from multiple patient characteristics to establish a prediction model. Specifically, the following variables were documented: age, gender, height, weight, alcohol use, history of liver diseases (hepatitis B, hepatitis C, and nonalcoholic fatty liver disease), admission, discharge, blood lipid level, antibiotics, other immunosuppressive agents, and Chinese traditional medicines. The scores of all risk factors were calculated using the machine learning method and represented by a ranking figure. The factor with a higher score had a more significant impact on the occurrence of hepatotoxicity. Factors with a score of zero were removed because they did not affect the prediction. A Shapley Additive Explanations (SHAP) figure demonstrated the positive or negative correlations between risk factors and hepatotoxicity. Risk factors with a large sample size would affect their impact on the SHAP figure. The SHAP figure was developed by using Python software (version 3.7).

Model development

Missing data were imputed using the missForest method, and variables with more than 30% missing data were discarded [35]. Patients included were randomly stratified (8:2) into the training set for modeling development and the testing set to evaluate the performance of the models. Using the selected risk factors as covariates, eight machine-learning models were established and analyzed using algorithms including eXtreme Gradient Boosting (XGBoost), AdaBoost, CatBoost, Gradient Boosting Decision Tree (GBDT), Light Gradient Boosting Machine (LightGBM), Tree-based Pipeline Optimization Tool (TPOT), Random Forest (RF), and Artificial Neural Network (ANN). The area under the curve (AUC) of the receiver operating characteristic (ROC) curve, representing the overall ability to classify and predict, is considered the primary metric for evaluating and comparing models. The accuracy, precision, sensitivity, specificity, recall, F1 scores, and average precision (AP) of precision–recall curve were also calculated. These metrics were used to assess the model performance comprehensively. The best-performing model was selected to establish a hepatotoxicity prediction model associated with low-dose MTX. The missForest method and machine learning models were developed and validated with open-source packages in Python software (version 3.7).

Statistical analysis

Categorical variables were summarized using frequency counts and percentages, and continuous variables were presented as means with standard deviations (SD) or medians with ranges. Comparisons between the training set and the testing set were made using the nonparametric Mann-Whitney U test for continuous variables and the χ2 test for categorical variables. Interrater reliability was assessed using Kappa's coefficient. By convention, P values of less than 0.05 were considered statistically significant. These analyses were performed using the SPSS 25.0 software.

Study Population

A total of 2,080 medical records were registered in the cohort during the study period, and 782 patients were enrolled in this study. The following patients were excluded: 171 had duplicate records, 588 were on high-dose MTX, and 539 had low-dose MTX as discharge medication (Fig. 1). Among the patients enrolled, the mean age was 47.85 ± 15.56 years (range 10 to 87 years), adult represented 97.31% (761/782), and the females represented 54.99% (430/782). The average body mass index (BMI) was 22.72 ± 3.91 kg/m² (range 13.27 to 41.14 kg/m²). All patients received oral MTX. A total of 279 (35.68%) patients experienced hepatotoxicity. The result of Kappa's coefficient was 0.96.Among these variables analyzed, BMI had 53 missing data points (6.78%) imputed using the missForest method. There was no significant difference between the processed data and the original data. The enrolled patients were divided into training and testing sets in a ratio of 8:2, with 625 and 157 patients, respectively. There were no significant differences in any variables between the training and testing sets (P > 0.05) (Table 1).

Table 1

The characteristics of patients
Variable	Total (n = 782)	Training set (n = 625)	Testing set (n = 157)	P
Hepatotoxicity
Yes	279	223	56	1.00
No	503	402	101	1.00
Gender
Male	352	281	71	1.00
Female	430	344	86	1.00
Age (years)	47.85 ± 15.56 (10–87)	47.67 ± 15.63 (14–78)	48.58 ± 15.33 (10–87)	0.40
Does of MTX (mg)	9.89 ± 2.12 2.5–20	9.80 ± 2.07 2.5–20	10.25 ± 2.27 5–20	0.28
First time taking MTX
Yes	501	400	101	1.00
No	272	225	56	1.00
Body mass index^* (kg/m²)	Original data
	729	586	143
	22.72 ± 3.91 (13.27–41.14)	22.78 ± 3.94 (13.27–41.14)	22.45 ± 3.77 (13.74–37.13)	0.71
	Processed data
	22.68 ± 3.81 (13.27–41.14)	22.77 ± 3.86 (13.27–41.14)	22.38 ± 3.23 (13.74–37.13)	0.37
Alcohol use
Yes	164	130	34	0.83
No	618	495	123	0.83
History of kidney disease
Yes	6	5	1	1.00
No	776	620	156	1.00
History of liver disease
Yes	32	23	9	0.26
No	750	602	148	0.26
Number of comorbidities	4.98 ± 2.97 (1–17)	4.90 ± 2.90 (1–17)	5.28 ± 3.23 (1–16)	0.69
Type 2 diabetes
Yes	69	59	10	0.27
No	713	566	147	0.27
Hyperlipidemia
Yes	41	30	11	0.32
No	741	595	146	0.32
Folate supplementation
Yes	723	575	148	0.40
No	59	50	9	0.40
Doses of folic acid/week	9.19 ± 3.33 (0–35)	9.16 ± 3.39 (0–35)	9.29 ± 3.07 (0–15)	0.05
NSAIDs use
Yes	276	231	45	0.06
No	506	394	112	0.06
Glucocorticoid use
Yes	441	350	91	0.72
No	341	275	66	0.72
Antibiotics use
Yes	153	117	36	0.26
No	629	508	121	0.26
Other immunosuppressive agent use
Yes	446	351	95	0.37
No	336	274	62	0.37
Number of medications	5.91 ± 2.93 (0–24)	5.97 ± 2.93 (0–24)	5.67 ± 2.92 (0–18)	0.92
Chinese traditional medicines use
Yes	68	56	12	0.75
No	714	569	145	0.75
*: The difference between processed data and original data was not statistically significant (P-values were 0.88, 0.94, and 0.87 in the total, training set, and testing set, respectively).
NSAID: non-steroidal anti-inflammatory drugs

Model performance

The visual comparisons of the eight models in the total population are shown in Fig. 2, including the precision-recall and the ROC curves. Random Forest achieved the highest AUC of 0.97, followed by XGboost (AUC = 0.94), Catboost (AUC = 0.91), LightGBM (AUC = 0.87), and TPOT (AUC = 0.78). The ROC curves of Adaboost, ANN, and GBDT were low, only 0.69, 0.65, and 0.53, respectively. The precision, accuracy, sensitivity, specificity, recall, and F1 values of the eight models are shown in Table 2.

Table 2

Model Performance
Models	Precision	Accuracy	Sensitivity	Specificity	F1
LightGBM	40.00%	59.87%	25.00%	20.79%	30.77%
GBDT	50.94%	59.24%	41.07%	30.69%	41.82%
Adaboost	51.35%	64.33%	33.93%	17.81%	40.86%
Catboost	42.86%	60.51%	32.14%	23.76%	36.73%
XGboost	43.18%	60.51%	33.93%	24.75%	38.00%
Random Forest	50.00%	64.33%	32.14%	17.82%	39.13%
TPOT	43.90%	61.15%	32.14%	22.77%	37.11%
ANN	36.36%	62.42%	7.14%	6.93%	11.94%

Adaboost and Random Forest had the highest accuracy (64.33%). Adaboost had the highest precision value (51.35%), followed by GBDT (50.94%). GBDT had the highest sensitivity value (41.07%), followed by Adaboost (33.93%) and XGboost (33.93%). GBDT had the highest specificity value (30.69%), followed by XGboost (24.75%). GBDT had the highest recall value (41.07%), followed by Adboost (33.93%) and XGboost (33.93%). GBDT had the highest F1 value (41.82%), followed by Adaboost (40.86%). These results showed that Adaboost had slight advantages in precision and accuracy with good recall, sensitivity, and F1 values. Adaboost had a significantly lower AUC than Random Forest (0.69 versus 0.97). After general consideration of the prediction performance, Random Forest was selected to predict the hepatotoxicity associated with low-dose MTX.

Hepatotoxicity and risk factors

A total of 279 patients experienced hepatotoxicity, with an incidence rate of 35.68%. The importance score ranking in the Random Forrest model is shown in Fig. 3. Importance scores were above zero for all risk factors, indicating that they had a greater or lesser impact on prediction. The top ten significant risk factors included BMI, age, number of drugs and comorbidities, doses of folic acid, antibiotic use, gender, immunosuppressive agents, taking MTX for the first time, and alcohol use, suggesting physicians should pay more attention to these factors and take the corresponding prevention measures. Among risk factors, the highest score was BMI (0.237), followed by age (0.198), number of drugs (0.151), and number of comorbidities (0.144), demonstrating their importance in hepatotoxicity associated with low-dose MTX.

The SHAP values of the risk factors are shown in Fig. 4. For risk factors including the number of comorbidities, the number of drugs, the use of antibiotics, male gender, the use of alcohol, infectious liver disease, dyslipidemia and the history of kidney disease, the color of the dot became redder as the SHAP value increased, and the color was bluer when the SHAP value decreased, showing the positive impact of these factors on the risk of hepatotoxicity. In contrast, risk factors, including BMI, taking MTX for the first time, and doses of folic acid, showed negative effects. Type 2 diabetes, other immunosuppressive agents, age, and Chinese traditional medicines showed unclear influence.

An effective prediction model is necessary to prevent the hepatotoxicity associated with low-dose MTX. In real-world studies, the variables are not independent but are related nonlinearly. Multivariate analysis methods are challenging for capturing complex relationships. Therefore, we innovatively attempted to apply machine-learning methods that can capture nonlinear relationships between variables. Machine learning can explore risk factors and establish a prediction model for hepatotoxicity associated with low-dose MTX through data learning. Our retrospective study analyzed 15 risk factors for hepatoxicity. The BMI with missing data was imputed using the missForest method, which has been shown to successfully handle missing values, particularly in data sets that include different variables [35]. The results did not show significant differences between the processed and original data.

The eight machine learning methods were applied to establish a prediction model. The results showed that these machine algorithms performed well, especially the Random Forest. Random Forest uses bootstrap aggregation of multiple regression trees to reduce the risk of overfitting and combine the predictions of many trees to produce more accurate predictions [36]. The Random Forest showed that its AUC was 0.97. The accuracy and precision were 64.33% and 50.00%, respectively. Both the recall and the F1 scores were satisfactory. Random Forest outperformed other models selected to build the prediction model for hepatotoxicity associated with low-dose MTX.

Analysis of risk factors showed that all 15 variables helped predict low-dose MTX-related hepatotoxicity. Among risk factors, BMI was considered the most critical risk factor, which had a negative relationship with hepatotoxicity, demonstrating that patients with lower BMI were more likely to experience hepatotoxicity. Therefore, the dose of MTX should be individualized based on height and weight to avoid hepatotoxicity and a dose reduction of MTX is appropriate for patients with low BMI. Male gender was also identified as an important risk factor in our study. However, the causal relationship between gender and hepatotoxicity associated with low-dose MTX remains controversial [37, 38] and requires further research.

The importance of the number of drugs, the number of comorbidities, and the use of antibiotics was also confirmed. As the primary organ for drug metabolism, the liver is more vulnerable to damage by drugs, active metabolites, or drug interactions [39, 40]. Multiple drug treatments and comorbid diseases can increase the risks of polypharmacy, drug interactions, and even medication errors [41, 42], increasing the risk of hepatotoxicity. Antibiotics are the most common cause of liver damage [43]. However, the potential for liver injury caused by antibacterial drugs was underestimated [44]. Several real-world studies showed that antibiotic-induced liver injury ranged from 13.5–65% [45–47]. Therefore, to avoid hepatotoxicity during MTX therapy, simplifying treatment regimens should be an important measure for the benefit of patients.

Alcohol consumption is well known to harm the liver, particularly in excess [48]. The American College of Rheumatology and the British Society of Rheumatology recommend limiting alcohol intake for patients on MTX treatment [49, 50]. Similarly, we found a positive relationship between alcohol use and hepatotoxicity associated with low-dose MTX. Although the importance score for alcohol consumption was not high in this study due to the relatively small number of patients who drank alcohol, we still recommend limiting or avoiding alcohol intake.

Supplementation with folic or folinic acid during MTX treatment can ameliorate ADEs. Worldwide guidelines currently support the coadministration of folic acid with MTX. The recommended doses range from 0.5 to 2 mg per day or 5mg per week [50–52]. In our study, the incidence of liver injury in patients taking folic acid was 34.85%, while the liver injury rate in 59 patients who did not take folic acid was up to 45.76%. However, the results indicated that patients taking 15-35mg/week folic acid had a higher risk of liver injury than those taking 5-10mg/week folic acid. This result may be related to the higher proportion of patients who drank alcohol in the folic acid high dose group. Therefore, we also recommend supplementation with folic acid during MTX treatment.

Metabolic syndrome is a biochemical and clinical condition characterized by visceral obesity, dyslipidemia, hyperglycemia, and hypertension [53]. Disorders associated with metabolic syndrome can be significant risk factors for fibrosis and progression of liver damage. Type 2 diabetes contributed to the biological processes that drove the severity of nonalcoholic fatty liver disease, which was the leading cause of developing chronic liver diseases [54]. Several studies showed that nonalcoholic steatohepatitis and hyperlipidemia contributed to MTX hepatotoxicity in patients with psoriasis [37, 55]. These were consistent with our results that type 2 diabetes and hyperlipidemia were significant risk factors for hepatotoxicity associated with low-dose MTX.

Hepatitis B and hepatitis C can cause liver damage, increasing the risk of liver toxicity and even liver fibrosis and cirrhosis in patients taking MTX [56]. Infectious liver disease was one of the important risk factors for hepatotoxicity in this study, while its importance score was not high. The reason might be that patients with infectious liver disease were only 4.1% of the study sample. For health and safety reasons in China, many physicians choose other alternative treatments for patients with infectious liver disease instead of MTX. Similarly, only six patients had a history of kidney disease in this study. Therefore, the importance score for the history of kidney disease was low.

Our study has the following limitations 1) knowledge about specific risk factors is still lacking in this study. Although factors such as taking MTX for the first time, other immunosuppressive agents, age, and Chinese traditional medicines affected the occurrence of hepatotoxicity, the direction of influence of these factors was unclear. These factors could be influenced by other factors, such as drug regimens (the number of drugs and drug interactions), gender, and BMI; 2) the sample size was small. Future studies should include more patient data from different health care centers; 3) long-term studies are required to verify the association of these risk factors with liver fibrosis or cirrhosis; 4) it was difficult to obtain relatively complete medication and examination data of outpatients, so only patients during hospitalization were included, which could lead to selection bias.

This novel study uses machine learning to establish the prediction model for low-dose hepatotoxicity associated with MTX. In the future, with the popularity of electronic patient record systems and the wide availability of structured patient data, this sophisticated computer algorithm will be implemented at the bedside to improve medication safety in patients taking methotrexate in clinical practice. However, due to the above limitations, further studies are required to test our findings.

Acknowledgments

The authors would like to thank Miye Wang, who works in the Information Center of the West China Hospital, for assistance with data extraction. We thank Jinsong Liu’s aid in data analysis.

Author Contribution

All the authors were involved in the study. Study design: QH, TX. Data extraction: QH, HW. Analysis and interpretation of data: QH. Writing – original draft: QH, HW. Writing – review & editing: QH, TX.

Funding

The National Key R&D Program of China supported this study, grant number 2020YFC2008302.

Availability of data and materials

All data generated or analyzed during this study are included in this published article.

Ethical approval

The ethics committee of the West China Hospital of Sichuan University, China, approved this study, approval number 2022-46 (date of approval January 13, 2022).

Consent to participate

The institutional ethics committee waived the requirement of informed consent for this retrospective study, and all data used in this study were fully anonymized.

Consent for publication

Not applicable.

Conflicts of interest

The authors have no conflicts of interest to declare.

West SG (1997) Methotrexate hepatotoxicity. Rheum Dis Clin North Am.23(4): 883–915. https://doi.org/10.1016/s0889-857x(05)70365-3
Farber S (1996) Chemotherapy in the treatment of leukemia and Wilm's tumor. JAMA. 198(8): 826–836.
Saag KG, Teng GG, Patkar NM, et al (2010) American college of rheumatology 2008 recommendations for the use of nonbiologic and biologic disease-modifying antirheumatic drugs in rheumatoid arthritis. Arthritis Rheum. 59(6): 762–784. https://doi.org/10.1002/art.23721
Cross M, Smith E, Hoy D, et al (2010) The global burden of rheumatoid arthritis: estimates from the Global Burden of Disease 2010 study. Ann Rheum Dis. 73(7):1316–22. https://doi.org/10.1136/annrheumdis-2013-204627.
Hunter TM, Boytsov NN, Zhang X, et al (2017) Prevalence of rheumatoid arthritis in the United States adult population in healthcare claims databases, 2004–2014. Rheumatol Int.37(9):1551–7. https://doi.org/10.1007/s00296-017-3726-1.
Myasoedova E, Crowson CS, Kremers HM, et al (2010) Is the incidence of rheumatoid arthritis rising?: Results from Olmsted County, Minnesota, 1955–2007. Arthritis Rheum.;62(6):1576–82. https://doi.org/10.1002/art.27425.
Icen M, Crowson CS, Mcevoy MT, et al (2009) Trends in incidence of adult-onset psoriasis over three decades: A population-based study. J Am Acad Dermatol. 60(3): 394–401. https://doi.org/10.1016/j.jaad.2008.10.062.
Michalek IM, Loring B, John SM (2017) A systematic review of worldwide epidemiology of psoriasis. J Eur Acad Dermatol Venereol,31(2):205–212. https://doi.org/10.1111/jdv.13854.
Lau CS, Chia F, Dans L, et al (2019) 2018 update of the APLAR recommendations for treatment of rheumatoid arthritis. Int J Rheum Dis. 22(3): 357–375. https://doi.org/10.1111/1756-185X.13513.
Kameda H, Fujii T, Nakajima A, et al (2019) Japan college of rheumatology guideline for the use of methotrexate in patients with rheumatoid arthritis. Mod Rheumatol. 29(1): 31–40. doi: 10.1080/14397595.2018.1472358.
Singh JA, Guyatt G, Ogdie A, et al (2019) Special Article: 2018 American college of rheumatology/national psoriasis foundation guideline for the treatment of psoriatic arthritis. Arthritis Care Res. 71(1):2–29. https://doi.org/10.1002/acr.23789.
Kuhn A, Aberer E, Bata-Csörgő Z, et al (2017) S2k guideline for treatment of cutaneous lupus erythematosus - guided by the european dermatology forum (edf) in cooperation with the european academy of dermatology and venereology (eadv). J Eur Acad Dermatol Venereol. 31(3),389–404. https://doi.org/10.1111/jdv.14053.
Nast A, Spuls PI, Kraaij G, et al (2017) European S3-Guideline on the systemic treatment of psoriasis vulgaris - Update Apremilast and Secukinumab - EDF in cooperation with EADV and IPC. J Eur Acad Dermatol Venereol. 31(12):1951–1963. https://doi.org/10.1111/jdv.14454.
Warris LT, Van Den Heuvel-Eibrink MM, Aarsen FK, et al (2016) Hydrocortisone as an intervention for dexamethasone-induced adverse effects in pediatric patients with acute lymphoblastic leukemia: results of a double-blind, randomized controlled trial. J Clin Oncol. 34(19): 2287–2293. https://doi.org/10.1200/JCO.2015.66.0761.
Matsuoka K, Kobayashi T, Ueno F, et al (2018) Evidence-based clinical practice guidelines for inflammatory bowel disease. J Gastroenterol. 53(3): 305–353. https://doi.org/10.1007/s00535-018-1439-1.
Ramanan AV, Dick AD, Jones AP, et al (2017) Adalimumab plus Methotrexate for Uveitis in Juvenile Idiopathic Arthritis. N Engl J Med. 376(17):1637–1646. https://doi.org/10.1056/NEJMoa1614160.
Van den Bosch F, Kruithof E, Baeten D, et al (2000) Effects of a loading dose regimen of three infusions of chimeric monoclonal antibody to tumour necrosis factor α (infliximab) in spondyloarthropathy: an open pilot study. Ann Rheum Dis. 59(6): 428–433. https://doi.org/10.1136/ard.59.6.428.
Mansouri B, Kivelevitch D, Campa M, et al (2016) Palmoplantar pustular psoriasis unresponsive to the interleukin-1β antagonist canakinumab. Clin Exp Dermatol. 41(3): 324–326. https://doi.org/10.1111/ced.12759.
Bai F, Li GG, Liu Q, et al (2019) Short-term efficacy and safety of IL-17, IL-12/23, and IL-23 inhibitors brodalumab, secukinumab, ixekizumab, ustekinumab, guselkumab, tildrakizumab, and risankizumab for the treatment of moderate to severe plaque psoriasis: a systematic review and network meta-analysis of randomized controlled trials. J Immunol Res. 10: 2546161. https://doi.org/10.1155/2019/2546161.
Kremer JM, Kaye GI, Kaye NW, et al (1995) Light and electron microscopic analysis of sequential liver biopsy samples from rheumatoid arthritis patients receiving long-term methotrexate therapy. Followup over long treatment intervals and correlation with clinical and laboratory variables. Arthritis Rheum. 38(9): 1194–1203. https://doi.org/10.1002/art.1780380904.
Curtis JR, Beukelman T, Onofrei A, et al (2010) Elevated liver enzyme tests among patients with rheumatoid arthritis or psoriatic arthritis treated with methotrexate and/or leflunomide. Ann Rheum Dis. 69: 43–7. https://doi.org/10.1136/ard.2008.101378.
Park SH, Choe JY, Kim SK (2010) Assessment of liver fibrosis by transient elastography in rheumatoid arthritis patients treated with methotrexate. Joint Bone Spine. 77:588–592. https://doi.org/10.1016/j.jbspin.2010.02.024.
Whiting-O'Keefe QE, Fye KH, Sack KD (1991) Methotrexate and histologic hepatic abnormalities: a meta-analysis. Am J Med. 90: 711–716.
Kalb RE, Strober B, Weinstein G, et al (2009) Methotrexate and psoriasis: National Psoriasis Foundation Consensus Conference. J Am Acad Dermatol. 60: 824–837. https://doi.org/10.1016/j.jaad.2008.11.906.
Clarey DD, Reid AT, Kiani R, et al (2021) Methotrexate Hepatotoxicity Monitoring Guidelines in Psoriasis and Rheumatoid Arthritis: Is There a Consensus? S D Med. 74(8): 363–366.
Jordan MI, Mitchell TM (2015) Machine learning: Trends, perspectives, and prospect. Science. 349(6245): 255–260. https://doi.org/10.1126/science.aaa8415.
Esteva A, Kuprel B, Novoa RA, et al (2017) Dermatologist-level classification of skin cancer with deep neural networks. Nature. 542(7639): 115–8. https://doi.org/10.1038/nature21056.
Gunčar G, Kukar M, Notar M, et al (2018) An application of machine learning to haematological diagnosis. Sci Rep. 8(1): 411. https://doi.org/10.1038/s41598-017-18564-8.
Qiu H, Yu HY, Wang LY, et al (2017) Electronic health record driven prediction for gestational diabetes mellitus in early pregnancy. Sci Rep. 7(1): 16417. https://doi.org/10.1038/s41598-017-16665-y.
Deo RC (2015) Machine learning in medicine. Circulation. 132: 1920–30. https://doi.org/10.1161/CIRCULATIONAHA.115.001593.
Goldstein BA, Navar AM, Carter RE (2017) Moving beyond regression techniques in cardiovascular risk prediction: applying machine learning to address analytic challenges. Eur Heart J. 38: 1805–14. https://doi.org/10.1093/eurheartj/ehw302.
Meyer A, Zverinski D, Pfahringer B, et al (2018) Machine learning for real-time prediction of complications in critical care: a retrospective study. Lancet Respir Med. 6: 905–14. https://doi.org/10.1016/S2213-2600(18)30300-X.
Mazaud C, Fardet L (2017) Relative risk of and determinants for adverse events of methotrexate prescribed at a low dose: a systematic review and meta-analysis of randomized placebo‐controlled trials. Br J Dermatol. 177(4): 978–86. https://doi.org/10.1111/bjd.15377.
Chalasani NP, Maddur H, Russo MW, et al (2021) ACG Clinical Guideline: Diagnosis and Management of Idiosyncratic Drug-Induced Liver Injury. Am J Gastroenterol. 116(5):878–898. https://doi.org/10.14309/ajg.0000000000001259.
Stekhoven DJ, Buhlmann P (2012) MissForest–nonparametric missing value imputation for mixed-type data. Bioinformatics. 28: 112–8. https://doi.org/10.1093/bioinformatics/btr597.
Breiman L (2001) Random forests. Mach Learn. 45(1): 5–32.
Chong MY, Chong VH, Earnest A, et al (2013) Prevalence and risk factors for methotrexate hepatoxicity in Asian patients with psoriasis. World J Hepatol. 5(5): 275–80. https://doi.org/10.4254/wjh.v5.i5.275.
Amital H, Arnson Y, Chodick G, et al (2009) Hepatotoxicity rates do not differ in patients with rheumatoid arthritis and psoriasis treated with methotrexate. Rheumatology. 48(9): 1107–10. https://doi.org/10.1093/rheumatology/kep176.
Sanoh S (2015)[In Vitro and in Vivo Assessments of Drug-induced Hepatotoxicity and Drug Metabolism in Humans]. Yakugaku. Zasshi. 135(11): 1273–9. https://doi.org/10.1248/yakushi.15-00200.
Ballet F (1997) Hepatotoxicity in drug development: detection, significance and solutions. J Hepatol. 2: 26–36. https://doi.org/10.1016/s0168-8278(97)80494-1.
Marcum ZA, Arbogast KL, Behrens MC, et al (2013) The utility of an adverse drug event trigger tool in veterans affairs nursing facilities. Consult Pharm. 28: 99–109. https://doi.org/10.4140/TCP.n.2013.99.
Hu Q, Qin Z, Zhan M, et al (2020) Validating the Chinese geriatric trigger tool and analysing adverse drug event associated risk factors in elderly Chinese patients: A retrospective review. PLoS One. 15(4): e0232095. https://doi.org/10.1371/journal.pone.0232095. eCollection 2020.
43., Einar SB (2017) Drug-induced liver injury due to antibiotics. Scand J Gastroenterol. 52(6–7): 617–623. https://doi.org/10.1080/00365521.2017.1291719.
Leitner JM, Graninger W, Thalhammer F (2010) Hepatotoxicity of antibacterials: Pathomechanisms and clinical. Infection. 38(1): 3–11. https://doi.org/10.1007/s15010-009-9179-z.
Mindikoglu AL, Magder LS, Regev A (2009) Outcome of liver transplantation for drug-induced acute liver failure in the United States: analysis of the United Network for Organ Sharing database. Liver Transpl. 15(7): 719–729. https://doi.org/10.1002/lt.21692.
Bjornsson E, Olsson R (2005) Outcome and prognostic markers in severe drug-induced liver disease. Hepatology. 42(2): 481–489. https://doi.org/10.1002/hep.20800.
Andrade RJ, Lucena MI, Fernandez MC, et al (2005) Drug-induced liver injury: an analysis of 461 incidences submitted to the Spanish registry over a 10-year period. Gastroenterology. 129 (2): 512–521. https://doi.org/10.1016/j.gastro.2005.05.006.
Warner A, Lunt M, Verstappen S (2017) Quantifying the hepatotoxic risk of alcohol consumption in patients with rheumatoid arthritis taking methotrexate. Ann Rheum Dis. 76(9): 1509–1514. https://doi.org/10.1136/annrheumdis-2017-211632.
Kremer JM, Alarcon GS, Lightfoot Jr RW, et al (1994) Methotrexate for rheumatoid arthritis. Suggested guidelines for monitoring liver toxicity. Arthritis Rheum. 37: 316–28. https://doi.org/10.1002/art.1780370304.
Chakravarty K, McDonald H, Pullar T, et al (2008) BSR/BHPR guideline for disease-modifying anti-rheumatic drug (DMARD) therapy in consultation with the British Association of Dermatologists. Rheumatology. 47: 924–5. https://doi.org/10.1093/rheumatology/ken322.
Griffith S, Fisher J, Clarke S, et al (2000) Do patients with rheumatoid arthritis established on methotrexate and folic acid 5 mg daily need to continue folic acid supplements long term? Rheumatology. 39: 1102–9. https://doi.org/10.1093/rheumatology/39.10.1102.
Shea B, Swinden MV, Ghogomu ET, et al (2014) Folic acid and folinic acid for reducing side effects in patients receiving methotrexate for rheumatoid arthritis. J Rheumatol. 41(6):1049–60. https://doi.org/10.3899/jrheum.130738.
Hurley BF, Hanson DE, Sheaff AK (2011) Strength training as a countermeasure to aging muscle and chronic disease. Sports Med. 41(4): 289–306. https://doi.org/10.2165/11585920-000000000-00000.
Kim H, Da SL, An TH, et al (2021) Metabolic Spectrum of Liver Failure in Type 2 Diabetes and Obesity: From NAFLD to NASH to HCC. Int J Mol Sci. 22(9): 4495. https://doi.org/10.3390/ijms22094495.
Langman G, Hall PM, Todd G (2001) Role of nonalcoholic steatohepatitis in methotrexate-induced liver injury. J Gastroenterol Hepatol. 16: 1395–1401. https://doi.org/10.1046/j.1440-1746.2001.02644.x.
Montaudié H, Sbidian E, Paul C, et al (2011) Methotrexate in psoriasis: a systematic review of treatment modalities, incidence, risk factors and monitoring of liver toxicity. J Eur Acad Dermatol Venereol. 25 Suppl 2: 12–18. https://doi.org/10.1111/j.1468-3083.2011.03991.x.

No competing interests reported.

Download PDF

Version 2

posted

You are reading this latest preprint version

Predicting hepatotoxicity associated with low-dose methotrexate using machine learning

Status:

Version 2

Abstract

Figures

Introduction

Methods

Study setting and the study population

Data extraction

Model development

Statistical analysis

Results

Study Population

Model performance

Hepatotoxicity and risk factors

Discussion

Conclusion

Declarations

References

Additional Declarations

Status:

Version 2