Drug repurposing for reducing the risk of cataract extraction in patients with diabetes mellitus: integration of artificial intelligence-based drug prediction and clinical corroboration

Abstract

Diabetes mellitus (DM) increases the incidence of age-related cataracts. Currently, no medication is approved or known to delay clinical cataract progression. Using a novel approach based on AI, we searched for drugs with potential cataract surgery-suppressing effects. We developed a drug discovery strategy that combines AI-based potential candidate prediction among 2650 Food and Drug Administration (FDA)-approved drugs with clinical corroboration leveraging multicenter electronic health records (EHRs) of approximately 800,000 cataract patients from the TriNetX platform. Among the top-10 AI-predicted repurposed candidate drugs, we identified three DM diagnostic ICD code groups, such as cataract patients with type 1 diabetes mellitus (T1DM), type 2 diabetes mellitus (T2DM), or hyperglycemia, and conducted retrospective cohort analyses to evaluate the efficacy of these candidate drugs in reducing the risk of cataract extraction. Aspirin, melatonin, and ibuprofen were associated with a reduced 5-, 10-, and 20-year cataract extraction risk in all types of diabetes. Acetylcysteine was associated with a reduced 5-, 10-, and 20-year cataract extraction risk in T2DM and hyperglycemia but not in T1DM patient groups. The suppressive effects of aspirin, acetylcysteine, and ibuprofen waned over time, while those of melatonin became stronger in both genders. Thus, the four repositioned drugs have the potential to delay cataract progression in both genders. All four drugs share the ability to directly or indirectly inhibit cyclooxygenase-2 (COX-2), an enzyme that is increased by multiple cataractogenic stimuli.

From In Silico Hypothesis to Validation: The Role of Real-World Evidence in the Preliminary Verification of AI-Generated Drug-Repositioning Candidates: A Comprehensive Review

“…The works of Shakibfar et al [ 39 ] (Crohn’s disease) and Bai et al [ 40 ] (Ulcerative Colitis) demonstrated that the analysis of medical registries allows for the exploratory verification of the impact of drugs on the risk of surgical interventions, such as intestinal resections. RWE showed similar precision in metabolic studies: the analyses of Koren et al and Brnabic et al [ 29 ] allowed for the identification of non-obvious therapeutic benefits of drugs used for other indications (e.g., alpha-1 receptor antagonists or dimethyl fumarate), and the study by Gao et al [ 30 ] revealed a protective influence of selected substances on the risk of developing cataracts in diabetics. Furthermore, the application of NLP methods by Patrick et al [ 41 ] in psoriasis proves that semantic literature analysis can effectively indicate candidates with anti-inflammatory effects.…”

“…Acetylcysteine was effective in type 2 diabetes (HR = 0.65) and hyperglycemia, but not in type 1 diabetes. The mechanism of action of these drugs is likely related to cyclooxygenase-2 (COX-2) inhibition and antioxidant activity [ 30 ]. A summary of the results of AI-driven drug repositioning in type 2 diabetes is presented in Table 4 .…”

“…In a study by Gao et al (2023), an integrated drug-repositioning strategy was proposed to reduce the risk of cataract surgery in patients with diabetes. Using the proprietary KG-Predict system based on knowledge graphs (integrating over 72,000 entities from databases such as DrugBank and GTEx), candidates were selected from among 2650 FDA-approved drugs.…”

From In Silico Hypothesis to Validation: The Role of Real-World Evidence in the Preliminary Verification of AI-Generated Drug-Repositioning Candidates: A Comprehensive Review

“…The works of Shakibfar et al [ 39 ] (Crohn’s disease) and Bai et al [ 40 ] (Ulcerative Colitis) demonstrated that the analysis of medical registries allows for the exploratory verification of the impact of drugs on the risk of surgical interventions, such as intestinal resections. RWE showed similar precision in metabolic studies: the analyses of Koren et al and Brnabic et al [ 29 ] allowed for the identification of non-obvious therapeutic benefits of drugs used for other indications (e.g., alpha-1 receptor antagonists or dimethyl fumarate), and the study by Gao et al [ 30 ] revealed a protective influence of selected substances on the risk of developing cataracts in diabetics. Furthermore, the application of NLP methods by Patrick et al [ 41 ] in psoriasis proves that semantic literature analysis can effectively indicate candidates with anti-inflammatory effects.…”

“…Acetylcysteine was effective in type 2 diabetes (HR = 0.65) and hyperglycemia, but not in type 1 diabetes. The mechanism of action of these drugs is likely related to cyclooxygenase-2 (COX-2) inhibition and antioxidant activity [ 30 ]. A summary of the results of AI-driven drug repositioning in type 2 diabetes is presented in Table 4 .…”

“…In a study by Gao et al (2023), an integrated drug-repositioning strategy was proposed to reduce the risk of cataract surgery in patients with diabetes. Using the proprietary KG-Predict system based on knowledge graphs (integrating over 72,000 entities from databases such as DrugBank and GTEx), candidates were selected from among 2650 FDA-approved drugs.…”

Association of adverse cardiovascular events with gabapentin and pregabalin among patients with fibromyalgia

“…This dataset encompasses a wide range of varied geographic regions, age groups, races and ethnicities, economic statuses, and types of health insurance. We previously used TriNetX platform to conduct retrospective cohort studies to investigate drug effects [9][10][11][12][13][14][15][16][17][18][19][20][21][22].…”

Machine Learning and Vision: Advancing the Frontiers of Diabetic Cataract Management

“…On their part, Crane et al (2023) found simulating actual cataract effects and generalizing results on the broader populations challenging. Gao et al (2023) [ 7 ] commented on the evolving set of genes that represented diabetic cataracts and, moreover, the fact that there was no detailed information about drug use in the EHR database. Hecht et al (2019) [ 9 ] commented that combining datasets from different sources may introduce biases in the diagnosis, while the small number of mixed cases reduces diagnosis potential.…”

From In Silico Hypothesis to Validation: The Role of Real-World Evidence in the Preliminary Verification of AI-Generated Drug-Repositioning Candidates: A Comprehensive Review

“…The works of Shakibfar et al [ 39 ] (Crohn’s disease) and Bai et al [ 40 ] (Ulcerative Colitis) demonstrated that the analysis of medical registries allows for the exploratory verification of the impact of drugs on the risk of surgical interventions, such as intestinal resections. RWE showed similar precision in metabolic studies: the analyses of Koren et al and Brnabic et al [ 29 ] allowed for the identification of non-obvious therapeutic benefits of drugs used for other indications (e.g., alpha-1 receptor antagonists or dimethyl fumarate), and the study by Gao et al [ 30 ] revealed a protective influence of selected substances on the risk of developing cataracts in diabetics. Furthermore, the application of NLP methods by Patrick et al [ 41 ] in psoriasis proves that semantic literature analysis can effectively indicate candidates with anti-inflammatory effects.…”

“…Acetylcysteine was effective in type 2 diabetes (HR = 0.65) and hyperglycemia, but not in type 1 diabetes. The mechanism of action of these drugs is likely related to cyclooxygenase-2 (COX-2) inhibition and antioxidant activity [ 30 ]. A summary of the results of AI-driven drug repositioning in type 2 diabetes is presented in Table 4 .…”

“…In a study by Gao et al (2023), an integrated drug-repositioning strategy was proposed to reduce the risk of cataract surgery in patients with diabetes. Using the proprietary KG-Predict system based on knowledge graphs (integrating over 72,000 entities from databases such as DrugBank and GTEx), candidates were selected from among 2650 FDA-approved drugs.…”

Association of adverse cardiovascular events with gabapentin and pregabalin among patients with fibromyalgia

“…This dataset encompasses a wide range of varied geographic regions, age groups, races and ethnicities, economic statuses, and types of health insurance. We previously used TriNetX platform to conduct retrospective cohort studies to investigate drug effects [9][10][11][12][13][14][15][16][17][18][19][20][21][22].…”

Machine Learning and Vision: Advancing the Frontiers of Diabetic Cataract Management

“…On their part, Crane et al (2023) found simulating actual cataract effects and generalizing results on the broader populations challenging. Gao et al (2023) [ 7 ] commented on the evolving set of genes that represented diabetic cataracts and, moreover, the fact that there was no detailed information about drug use in the EHR database. Hecht et al (2019) [ 9 ] commented that combining datasets from different sources may introduce biases in the diagnosis, while the small number of mixed cases reduces diagnosis potential.…”

From In Silico Hypothesis to Validation: The Role of Real-World Evidence in the Preliminary Verification of AI-Generated Drug-Repositioning Candidates: A Comprehensive Review

“…The works of Shakibfar et al [ 39 ] (Crohn’s disease) and Bai et al [ 40 ] (Ulcerative Colitis) demonstrated that the analysis of medical registries allows for the exploratory verification of the impact of drugs on the risk of surgical interventions, such as intestinal resections. RWE showed similar precision in metabolic studies: the analyses of Koren et al and Brnabic et al [ 29 ] allowed for the identification of non-obvious therapeutic benefits of drugs used for other indications (e.g., alpha-1 receptor antagonists or dimethyl fumarate), and the study by Gao et al [ 30 ] revealed a protective influence of selected substances on the risk of developing cataracts in diabetics. Furthermore, the application of NLP methods by Patrick et al [ 41 ] in psoriasis proves that semantic literature analysis can effectively indicate candidates with anti-inflammatory effects.…”

“…Acetylcysteine was effective in type 2 diabetes (HR = 0.65) and hyperglycemia, but not in type 1 diabetes. The mechanism of action of these drugs is likely related to cyclooxygenase-2 (COX-2) inhibition and antioxidant activity [ 30 ]. A summary of the results of AI-driven drug repositioning in type 2 diabetes is presented in Table 4 .…”

“…In a study by Gao et al (2023), an integrated drug-repositioning strategy was proposed to reduce the risk of cataract surgery in patients with diabetes. Using the proprietary KG-Predict system based on knowledge graphs (integrating over 72,000 entities from databases such as DrugBank and GTEx), candidates were selected from among 2650 FDA-approved drugs.…”

Association of adverse cardiovascular events with gabapentin and pregabalin among patients with fibromyalgia

“…This dataset encompasses a wide range of varied geographic regions, age groups, races and ethnicities, economic statuses, and types of health insurance. We previously used TriNetX platform to conduct retrospective cohort studies to investigate drug effects [9][10][11][12][13][14][15][16][17][18][19][20][21][22].…”

Machine Learning and Vision: Advancing the Frontiers of Diabetic Cataract Management

“…On their part, Crane et al (2023) found simulating actual cataract effects and generalizing results on the broader populations challenging. Gao et al (2023) [ 7 ] commented on the evolving set of genes that represented diabetic cataracts and, moreover, the fact that there was no detailed information about drug use in the EHR database. Hecht et al (2019) [ 9 ] commented that combining datasets from different sources may introduce biases in the diagnosis, while the small number of mixed cases reduces diagnosis potential.…”