Artificial intelligence in the field of pharmacy practice: A literature review

Artificial intelligence (AI) is a transformative technology used in various industrial sectors including healthcare. In pharmacy practice, AI has the potential to significantly improve medication management and patient care. This review explores various AI applications in the field of pharmacy practice. The incorporation of AI technologies provides pharmacists with tools and systems that help them make accurate and evidence-based clinical decisions. By using AI algorithms and Machine Learning, pharmacists can analyze a large volume of patient data, including medical records, laboratory results, and medication profiles, aiding them in identifying potential drug-drug interactions, assessing the safety and efficacy of medicines, and making informed recommendations tailored to individual patient requirements. Various AI models have been developed to predict and detect adverse drug events, assist clinical decision support systems with medication-related decisions, automate dispensing processes in community pharmacies, optimize medication dosages, detect drug-drug interactions, improve adherence through smart technologies, detect and prevent medication errors, provide medication therapy management services, and support telemedicine initiatives. By incorporating AI into clinical practice, health care professionals can augment their decision-making processes and provide patients with personalized care. AI allows for greater collaboration between different healthcare services provided to a single patient. For patients, AI may be a useful tool for providing guidance on how and when to take a medication, aiding in patient education, and promoting medication adherence and AI may be used to know how and where to obtain the most cost-effective healthcare and how best to communicate with healthcare professionals, optimize the health monitoring using wearables devices, provide everyday lifestyle and health guidance, and integrate diet and exercise.


Introduction
Alan Turing's seminal work, "Computing Machinery and Intelligence," published in 1950, marked the beginning of the artificial intelligence (AI) debate. 1 In 2004, John McCarthy defined AI as "the science and engineering of making intelligent machines, especially intelligent computer programs.". 2 AI has emerged as a transformative technology that has revolutionized a wide range of industries worldwide.From finance to healthcare, manufacturing, and transportation, AI has been at the forefront of innovation, enabling previously inconceivable advances.AI has paved the way for unprecedented automation, efficiency, and decision-making capabilities by leveraging intelligent algorithms, machine learning (ML), and data analytics.

Artificial intelligence in healthcare
AI in healthcare has evolved dramatically over the last five decades, leading to significant advancements in a variety of medical fields. 3The introduction of ML and deep learning (DL) has expanded AI applications, enabling personalized medicine rather than relying solely on algorithms.AI has significantly impacted clinical decision-making, disease diagnosis, as well as clinical, diagnostic, rehabilitative, surgical, and predictive practices. 4his advancement in AI technology has paved the way for improved diagnostic accuracy, streamlined provider workflow, improved clinical operation efficiency, disease, and therapeutic monitoring, precise procedures, and, ultimately, better patient outcomes. 5,6. 2

. Pharmacy practice
Pharmacy practice is an integral part of the healthcare system, which ensures safe and effective medication management and optimized patient care, through various activities such as medication reconciliation, medication review, medication therapy management (MTM), providing drug information, patient education, adverse drug reaction (ADR) monitoring and interprofessional collaborations. 7ith rapid advancements in the healthcare sector, the number of prescriptions, complex drug regimens, and administrative tasks has increased noticeably.As a result, there is an increasing demand for advanced technological solutions that can assist healthcare professionals in their daily responsibilities and optimize healthcare service delivery. 8he incorporation of AI technologies provides pharmacists with tools and systems that help them make accurate and evidence-based clinical decisions.By using AI algorithms and ML, pharmacists can quickly analyze large amounts of patient data, including medical records, lab results and medication profiles.This allows them to identify potential drug-drug interactions, assess the safety and efficacy of medicines, and make informed recommendations tailored to individual patients. 3,8,9he application of AI in various areas within the field of pharmacy practice has shown promising prospects.However, existing research gaps need to be addressed to harness the complete potential of AI technologies.The most important aspect is the comprehensive implementation of AI services within existing pharmacy systems and understanding its impact on health and economic outcomes.In this review, we will be exploring the various AI applications in the field of pharmacy practice; the research gaps and challenges; and highlighting the future directions for research within the field.

Methods
To identify topics of interest for this narrative review, the various databases (PubMed, Google Scholar, and Scopus) were searched for relevant articles.Various search terms were used to identify the relevant literature, which included "Artificial intelligence," Adverse drug reaction," "ADR," "Machine learning," "Deep learning," "Neural networks," "Clinical decision support systems," "Medical Order Entry Systems," "Computerized Provider Order Entry," "Pharmacy practice," "Clinical pharmacy," "Community pharmacy," "Hospital pharmacy," "Pharmacist," "Medication therapy management," "Drug dispensing," "Medication reconciliation," "Medication adherence," "Medication optimization," "Pharmaceutical care," "Precision medicine."The reference list of the relevant articles was also reviewed to identify potentially important papers pertaining to the topic.Two authors independently conducted the search and the most appropriate ones were included into the review.

Adverse drug reaction (ADR) detection
AI has been utilized in several studies for ADR prediction and detection.One such study conducted by Mohsen and colleagues, which combined two distinct datasets: drug-induced gene expression profiles from the Open Toxicogenomics Project-Genomics Assisted Toxicity Evaluation Systems (TG-GATEs) database and ADR occurrence data from the Food and Drug Administration (FDA) Adverse Events Reporting System (FAERS) database in conjunction with Deep Neural Networks (DNN) for ADR prediction.It includes data filtering and cleaning, feature selection, and hyperparameter tuning. 10alçn et al. developed a ML-based clinical decision support tool (risk score) that predicts whether the identified ADRs would occur by integrating the severity with neonatal adverse event severity scale (NAESS) and probability with the 'Du'ADRs algorithm into the risk matrix analysis performed by a multidisciplinary team that included a clinical pharmacist.Decision tree induction, a ML method, was used by Hammann et al. to determine the chemical, physical, and structural properties of compounds that predispose them to cause ADRs.For allergic, renal, CNS, and hepatic ADRs, the models had high predictive accuracies (78.9-90.2%). 11,12n a study by Cami et al., a logistic regression classifier to predict unknown ADRs for marketed drugs using structural properties of the drug-ADR network as well as chemical and taxonomic properties of drugs as features was developed. 13Rahmani et al. used a random walk algorithm to predict unknown ADRs in a network with drug and ADR nodes, where drug-ADR edges represent known ADRs and drug-drug edges indicate drug target similarity, but they did not validate new ADRs in any real-world clinical data. 14Bresso et al. also created a database of the drug, ADR, and target knowledge and used decision trees and inductive logic programming to predict ADR profiles (rather than individual ADRs), which they validated using FAERS. 15ean et al. created a knowledge graph with four different types of nodes: drugs, protein targets, indications, and adverse reactions.Using this graph, they created a ML algorithm based on a simple enrichment test and demonstrated how well this method performs at classifying known causes of adverse reactions. 16][19][20][21][22][23] Overall, these studies highlight the broad range of AI applications in ADR detection, involving prediction models to clinical decision support tools and knowledge graph-based algorithms.

Clinical decision support system (CDSS)
A clinical decision support system (CDSS) is designed to improve healthcare delivery by supplementing medical decisions with targeted clinical knowledge, patient information, and other health data.Individual patient characteristics are matched to a computerized clinical knowledge base in a CDSS, and patient-specific assessments or recommendations are then presented to the clinician for a decision.This technology enables pharmacists to sift through data and intervene to prevent medication errors, reduce patient complications, and save money. 24,25

Community Pharmacy
Healthcare systems are rapidly transitioning from a single hospitalbased care module to a collaborative care system based in the community.Pharmacists can help to improve patient safety and efficacy of pharmacotherapy from the hospital to the community.The "robotic dispensing system" in the community pharmacies prepares prescribed medicines.It consists of three parts 26 :.
(1) An automated dispensing robot operated by pharmacy support staff, (2) An automated dispensing robot for powdered medication, and (3) A bar-coded medication dispensing support system with personal digital assistance.
ML models also allow e-mails to be personalized faster and more accurately than any human.Chatbots can be used to improve service delivery efficiency. 8Chatbots can simulate interactions between customers and customer service representatives.Chatbots can automatically resolve customer complaints and queries, and difficult questions are routed to human staff.Chatbots in community pharmacies can be programmed to simulate interactions between pharmacists and patients. 27algreen collaborated with a telehealth company, to develop a video chat platform for patients to interact with healthcare professionals. 28AI can also help with inventory management, where community pharmacists can predict what their patients will require in the future, stock them, and use personalized software to send e-mails to remind patients of drug requirements.A patient's future drug purchase can be predicted using AI-powered data analytics.The pharmacist will be able to make better stock procurement decisions if AI can predict the patient's drug purchase. 8,29n AI company, created a software for a German online and catalog retailer, which can predict what the retailer will sell in 30 days with 95% accuracy.This resulted in reduction in delivery schedule for purchased products from one to two days by allowing direct delivery from the supplier to the consumer without passing through the warehouse. 30he University of California San Francisco (UCSF) Medical Center also prepares and tracks medications using robotic technology.They claim that the technology has prepared 3,500,000 medication doses without error.The robot has proven to be far superior to humans in terms of both size and ability to deliver accurate medications.The robotic technology's capabilities include the preparation of oral and injectable medicines, including toxic chemotherapy drugs.The robotics package, and dispense individual doses of pills.The machines also assemble the doses onto a bar-coded plastic ring, which contains all medications that a patient must take within 12 h.The automated system's capabilities include the ability to prepare sterile preparations for chemotherapy as well as fill intravascular syringes with the appropriate medications. 31

Computerized prescriber order entry (CPOE)
Medication errors, according to the Institute of Medicine, are the most common type of error in healthcare, accounting for approximately 7000 deaths each year. 32Although there are numerous causes of medication errors, published research estimates that 11.4% of these errors are directly related to drug name mismatches, such as illegible prescriptions, confusing dosage forms, and misunderstood abbreviations. 33omputerized Physician Order Entry (CPOE), also known as Computerized Provider Order Entry or Computerized Practitioner Order Entry, is a process by which a physician enters and sends medication orders and treatment orders, as well as laboratory, admission, radiology, referral, and procedure orders electronically through a computer application, rather than using traditional methods such as paper charts, verbal orders, telephone, and fax.This method reduces errors caused by illegible handwriting or transcription errors in medication instructions.34 These CPOE systems control the selection, display, and storage of medication histories and the electronic transmission of medication orders to dispensing pharmacists and pharmacies.This new paradigm offers numerous opportunities to protect patient safety (e.g., allergy or renal dosing alerts), but also raises the possibility of many new types of predictable and unpredictable prescribing and dispensing errors.35

Dose recommendations
Patients can benefit from a personalized AI/ ML-based dosage recommendation system that incorporates data from multiple sources, such as safety and effectiveness metrics, electronic health records, disease details, treatment history, and patient feedback.These systems aim to improve treatment efficacy while minimizing side effects.Reinforcement learning algorithms have shown promise in predicting and adjusting dosages for precision-based cancer treatment. 36he most recent innovation with the potential to improve chronic disease care is a novel dosing optimization system, which is a platform for actionable dosing optimization that was created to improve chemotherapy dosing precision.The algorithm considers treatment response over time, predicting dosing requirements dynamically to maintain required efficacy and safety levels. 371.5.1.AI in high-risk drug dosing.Because of the dynamic profile of patients receiving the drug, optimizing vancomycin therapy remains a challenge in current clinical practice.Many factors, including renal function, concomitant drugs, and weight, are known to influence vancomycin dose-concentration response.Various approaches, such as dosing nomograms and Bayesian estimation methods, have been used in clinical practice to guide clinicians in vancomycin dosing.38,39 Wang Z, Ong CL, and Fu Z created a new AI-assisted dosage titration approach that has the potential to improve on traditional approaches.This approach is especially useful for guiding decision-making for inexperienced doctors in making consistent and safe dosing recommendations for high-risk medications like vancomycin.39 Researchers have also developed prediction models for the dosage of drugs like digoxin 40 and warfarin, 41 aiding in avoiding ADEs from dosage errors.

Drug-drug interactions
Drug-drug interactions (DDIs) have been identified as a significant cause of ADRs, which contribute to rising healthcare costs. 42,43Predicting DDI necessitates the use of multiple drug characteristics and known DDI.The most used databases are DrugBank, 44 SIDER, 45 TWO-SIDES, 46 Kyoto Encyclopedia of Genes and Genomes (KEGG), 47 Lexicomp, 48 and Micromedex. 49xisting DDI computational methods are classified into three types: similarity-based methods, networks-based methods, and ML methods.Van Laere et al. developed an algorithm that predicts QTc prolongation and issues alerts when DDIs increase the risk of QTc prolongation.Suyu Mei and Kun Zhang proposed a simple f-drug target profile representation to depict drugs and drug pairs, which was used to build an l2-regularized logistic regression model to predict DDIs. 43ong et al. created a largescale DDI predictor by combining five types of drug similarities: 2D molecular structure similarity, 3D pharmacophoric similarity, drug interaction profile similarity, target similarity, and adverse effect similarity, and provided a Polynomial Kernel Support Vector Machines (PK-SVM) classifier to carry out the predictive work.

Electronic health record (EHR)
The implementation of a new predictive EHR algorithm can lead to improved clinical decisions through software can detect and alert, when a prescribed drug appears to deviate from its pattern of appropriate use by using large amounts of EHR data and AI to learn patterns concerning appropriate medication use.Furthermore, AI could aid in drug selection decisions by indicating which patients are unlikely to experience adverse effects from a specific drug via automated classification. 52,53atient Safety Learning Laboratory (PSLL) embedded AI into the EHR systems can identify, assess, and mitigate threats to patient safety. 54he use of natural language processing (NLP) and ML in hospital and health system pharmacies to access and analyze unstructured, free-text information captured in millions of EHRs (e.g., medication safety, patients' medication history, adverse drug reactions, interactions, medication errors, therapeutic outcomes, and pharmacokinetic consultations) may become an essential tool to improve patient care and perform real-time evaluations of the efficacy of medications.This strategy has enormous potential to support risk-sharing agreements and guide decision-making in pharmacy and therapeutics (P&T) Committees. 55imilar model was developed by Balestra M et al., a predictive model for flagging orders requiring intervention using only information about the ordering provider's interaction with the EHR. 56

Identification of potentially inappropriate drug
Potentially inappropriate medications (PIMs) are medications whose risks outweigh the benefits when administered to patients. 57The S.H. Chalasani et al.  prevalence of comorbid conditions and polypharmacy among elderly patients puts them at risk of potentially inappropriate prescribing (PIP).There are currently several criteria for assessing PIP, including the Beers criteria 58 and the STOPP/START criteria. 59Despite the fact that these criteria are widely used for post-event evaluation.However, by detecting PIP early, physicians and pharmacists will be able to identify patients at risk of PIP and implement individualised interventions to reduce the risk of ADR.Several AL/ML algorithms are increasingly being used to develop predictive models for PIMs prescription. 60hun-Tien Tai and colleagues predicted the risk of digoxin treatment using ML.The results demonstrated that the best model performance successfully identified the risk.This study found that ML techniques can improve prediction accuracy for high alert drug (HAD) medication treatment, lowering the risk of ADEs, and improving medication safety. 61Wongyikul et al. created a HAD screening protocol with a ML model that used Gradient Boosting Classifier and screening parameters to identify HAD prescription errors from outpatient and inpatient drug prescriptions.The ML algorithm identified over 98% of actual HAD mismatches in the test set and 99% in the evaluation set when screening drug prescription events with a risk of HAD inappropriate use.This study demonstrated that ML played an important role in screening and reducing errors in HAD prescriptions. 62atel et al. developed predictive models using ML algorithms to identify predictors of inappropriate use of nonsteroidal antiinflammatory drugs (NSAIDs) of PIP in elderly patients with osteoarthritis. 63ingwei et al. used five sampling methods, three feature screening methods, and eighteen ML algorithms to handle process data and establish risk warning models for potentially inappropriate prescriptions for elderly patients with cardiovascular disease.The study enrolled 404 patients, 318 (78.7%) with PIP, 112 (27.7%) with PIMs rate, and 273 (67.6%) with potential prescribing omissions errors (PPO).Following data sampling and feature selection selecting characteristics, 15 datasets were obtained, based on which 270 risk warning models were built to predict PIP, PPO, and PIM, respectively.The study results found the important factors in the PIP risk warning model to be angina, the number of drugs, the number of diseases, and age.The risk warning platform built was able to predict PIP, PIM, and PPO with acceptable accuracy, predictive performance, and clinical application potential. 60

Medication adherence
Approximately half of patients with chronic diseases do not take their medications as prescribed, resulting in increased morbidity and mortality; and costing an estimated 100 billion USD per year. 64lthough pharmacist-led interventions appear to be the most effective in promoting medication adherence, they are frequently complex, involving multiple healthcare providers and multiple components.Since medication adherence barriers are complex and varied, solutions to improve adherence must be multifactorial, and AI technology may be viewed as a promising aspect of such interventions. 65here are various AI technologies used for promoting and monitoring medication adherence.Based on their technical designs and adherence monitoring functions, the identified technology types were divided into eight major groups: electronic pillboxes or bags, electronic pill bottles, ingestible sensors, blister pack technology, electronic medication management systems, patient self-report-based technology, video-based technology, and motion sensor technology.

The medication event monitoring system (MEMS).
A sensor embedded in the pill cap allows the MEMS to record every time the patient opens the pill bottle.Some newer electronic pill bottle technologies can wirelessly transmit patient medication adherence data, allowing for real-time assessment and monitoring of patient medication adherence. 66earfield communication (NFC) capabilities are frequently built into newer smartphones and medical devices, which can simplify the workflow of patient self-monitoring.NFC is a short-range communication standard that allows data transmission between two NFC devices within a few centimeters (touching).NFC tags can be used to track medication adherence.Patients can track their medication intake by bringing such NFC tags into contact with a smartphone. 67pecial blisters can be used to track medication intake via NFC.These smart blisters are protected by a foil that contains an electronic circuit.When the tablets or capsules are removed from the blister, a microcontroller detects the interruptions in the conductive paths and records the time and date. 67,68eDispensers, which can both remind patients to take their medications and directly provide them with them. 691.9.2.Motion sensor technology.Other methods are using triaxial accelerometers in wireless wearable devices to record and analyze the patient's hand movements.The addition of a fluorophore to the medication, which can be detected in the bloodstream with a monitoring device on the patient's wrist.70,71 Ingestible sensors, also referred to as digital pills or digital ingestion monitoring, are a technological system that consists of microsensors, an adhesive external monitor worn on the abdomen, and a mobile app.The medication and micro-ingestible sensors are co-encapsulated and ingested into the body, where stomach gastric fluids dissolve the capsule containing the medication and sensor.When the sensor detects gastric fluid, it sends a unique signal to the external monitor.The detected ingestion event is sent to a mobile app, which uploads the event's date and time stamp, as well as other recorded physiological measures (for example, heartbeat), to a central server.66

Electronic medication management systems (EMMS).
[74] 3.1.9.4.Video-based monitoring technology.Most video-based adherence monitoring technologies use video cameras to allow patients to self-record medication ingestion event videos, which are retrospectively analyzed by HCPs or, AI.Patient Self-reporting Technology, like EMMS, differ in their specific functionalities, but they all collect subjective medication adherence data by interacting with the patient via phone calls, smart buttons, eDiaries, web-based platforms, and mobile apps.For most self-reported devices, patient adherence is available in realtime. 66

Medication errors identification
The Food and Drug Administration (FDA) receives over 100,000 reports from the United States each year regarding suspected medication errors (MEs). 75Prescription errors occur at rates ranging from 0.3 to 9.1% in European hospitals, while dispensing errors occur at rates ranging from 1.6 to 2.1%. 76According to reports, a comprehensive and systematic approaches to patient safety can prevent up to 70.2% of MErelated harm.Implementation of electronic prescription systems, robust medication error surveillance, and the use of barcode medication administration systems are promising strategies for reducing MEs occurrence. 77n Israel based company was first to launch a commercial system that uses ML techniques to prevent prescription errors.This system detects overdose and underdose prescriptions with low false-positive rates by analyzing EHRs and generating automatic alerts. 78Segal et al. evaluated the utility of a ML-based CDSS in clinical practice.The system examined 78,017 prescriptions, generated 282 alerts (0.4%), and resulted in the discontinuation or modification of 135 prescriptions. 79antos H. et al. proposed an unsupervised method for detecting potential outlier prescriptions called density-distance-centrality (DDC).A dataset of 563 thousand prescribed medications was used to compare the proposed approach to various state-of-the-art outlier detection techniques.In comparison to other methods used to solve this problem, the approach achieves better results in the task of detecting overdose and underdose in medical prescriptions in the experiments.Furthermore, most of the false positives detected by the algorithm were potential prescription errors. 80A software as a service (SaaS) system that uses AI to assist clinical pharmacists in decision-making, was developed to improve patient outcomes. 81Nagata et al. used ML to create an algorithm for detecting prescription errors in overdoses and underdoses. 82imilarly, Yalçin N. et al. developed a model that predicts MEs detected by the clinical pharmacist during the pharmacotherapy process (prescription, preparation, administration, and monitoring) of patients admitted to the NICU using a newborn-centered approach (ML algorithms).The goal was to reduce physician and nurse workload while preventing MEs as part of pharmacotherapy optimization. 83 French company launched a hybrid AI decision support system in a typical hospital setting, which combined ML and a rule-based expert system to predict medication errors at the patient level rather than at the level of individual prescription orders.84

Medication therapy management (MTM)
The comprehensive medication management (CMM)-Wrap program used a novel AI platform that combines population health and telemedicine to identify and prioritize at-risk members and provide AI decision support for interventions using robust data collection and reporting as well as proprietary MedRiskScores (risk scores).This CMM-Warp involved a disease therapy management provider, combining population health and telemedicine to identify and prioritize the patients with increased risk.They provided remote telephonic services by teams of disease management-trained medical assistants and clinical pharmacists.The research results shown that when pharmacists and medical assistants who have received appropriate training work together with advanced AI systems to deliver CMM services over the phone, led to a decrease in healthcare expenses and a reduction in emergency department visits and hospital admissions.These positive outcomes can be considered potential signs of enhanced well-being. 85uring the COVID-19 pandemic, a grade 3 A specialized hospital in Shanghai, launched an AI-based internet hospital pharmacy service. 86he prescription rules were developed and embedded into the internet hospital system to review the prescriptions using AI, after which the pharmacists would review and the medications would be dispensed after a double check.Then, a "medicine pick-up code" is generated, which is a Quick Response (QR) code that represents a specific offline self-pick-up order (fragile drugs, high-risk drugs, and drugs requiring special management and storage at 2-8 • C).Other drugs that could be delivered were entrusted to a third-party pharmaceutical company.Patients or volunteers could retrieve medications from an offline hospital or drugstore by scanning the QR code through the window and waiting for the dispensing machine or pharmacist to dispense the drugs.They also provided medication consultation services, where a volunteer team of licensed pharmacists with extensive clinical experience provided free medication consultation services online. 86

Telehealth
Telehealth, also known as telemedicine, is the use of medical information exchanged between sites via electronic communication to improve health outcomes. 87hatbots can speed up and simplify history taking by using NLP to provide prompts and questions based on patient responses, such as selfreporting symptoms, and can provide possible diagnoses, including ADE detection, that can be coded and applied to future patient visits. 88 conversational AI platform that complies with the Health Insurance Portability and Accountability Act, developed an adverse event (AE) detection module that uses deep learning and NLP via a virtual assistant to recognize and differentiate between different AEs based on the questions and phrases presented.Once the AE has been identified, the module will automatically transcribe and export the information to the pharmaceutical company, as well as assist with FDA reporting.89 In telehealth settings, AI has the potential to improve pharmacovigilance.One study found that using automated phone calls to contact patients starting new medications helped to identify ADEs.Patients whose responses indicated the possibility of ADEs were referred to a pharmacist for further assessment.AI could be used to predict which patients should be screened and when they should be contacted.This, in conjunction with other technologies such as patient portals and texting, has the potential to improve the efficiency and effectiveness of pharmacovigilance efforts.90 Patients benefit from health information technologies such as telemonitoring, mobile health applications, and wireless monitoring devices.Monitoring data, disease information, symptom diaries, medication logs, reminders, nutrition diaries, and communication tools are examples of these.Wearable devices and mobile health apps can monitor personal analytics, physical status, and physiological parameters, which can help with medication schedules.Patients use networked medical devices ranging from consumer products such as Fitbit and Apple Watch to wearable external devices such as portable insulin pumps and internally embedded devices such as pacemakers.Providers can assess real-time dynamic data generated by wearable devices using software applications on various devices.88,91 The summary of findings is shown in Table 1.

Data privacy and security
Concerns about data privacy and security have arisen with the widespread use of AI-based applications.Health information is sensitive and a common target for data breaches.Patient data protection is thus critical. 95Some patients may be concerned that their data collection will infringe on their privacy, and lawsuits have been filed in response to data-sharing between large health systems and AI developers. 96Patient consent is an important factor in data privacy concerns, as healthcare organizations may allow the large-scale use of patient data for AI training without obtaining sufficient individual patient consent.Deep-Mind Health was acquired by Google in 2018.Their application, Streams, which contains an algorithm for managing patients with acute kidney injuries, was making headlines after it was revealed that the National Health Services (NHS) had given DeepMind servers the data of 1.6 million patients in order to train its algorithm without the patient consent. 94,97

Bias
Biases in the data collection used to train AI models can lead to biased results. 98Minorities, for example, may be under-represented in datasets due to racial biases in dataset creation, resulting in lower-thanexpected prediction performance.Even if AI systems are trained on accurate, representative data, problems may arise if the data reflects underlying biases and inequalities in the health-care system. 92For example, African-American patients receive less opioid analgesia on average than white patients; an AI system learning from health-care records may learn to recommend lower doses of opioid analgesia to African-American patients, despite the fact that this decision is based on systemic bias rather than biological reality. 99

Data integration
Following the acquisition of data, the next challenge is the development of AI technology.Overfitting can occur when the system learns irrelevant relationships between patient variables and outcomes.It is caused by having too many variable parameters in relation to outcomes, and as a result, the algorithm predicts using inappropriate features. 100ome classification and clustering algorithms may produce very good accuracy when applied to a small amount of data; however, this may not be realistic or applicable.To be used in AI techniques, the   To develop a natural language processing (NLP) system that will identify mentions of symptoms and drugs in clinical notes and label the relationship between the mentions as indications or adverse drug effects (ADEs).
Database study -Named Entity Recognition (NER) Clinical notes from the UMASS hospital were de-identified and manually annotated into categories.
The NLP system validation was carried out against the evaluation set provided by the MADE 1.0 challenge, and the performance of our system was compared to that of other submitted systems.The micro-averaged F1 score for NER was 80.9%, RE was 88.1%, and the final system was 61.2%.Duan L, 2012 22 USA To develop methods for identifying the associations that the observational medical outcomes partnership (OMOP) defined in order to simulate data from the observational simulated dataset.

Database study
The simulated dataset contains ten million people, 90 million drug exposures from 5000 different drugs, and 300 million condition occurrences from 4500 different conditions from a period of over a ten-years.
The experimental results show that the proposed pattern discovery method improves the standard baseline algorithm-chisquare-by 23.83% on the simulated OMOP dataset.

China
Based on potential drug perturbations on associated genes and signaling pathways, an attempt was made to simplify computational modelling for drug-drug interaction prediction.

Database study
Only drugs that have been discovered to target at least one human gene were represented in the drug target profile.
The SP, SE, and MCC metrics on the two classes show that the proposed framework is less biased, with 0.9556 on the positive class, 0.9402 on the negative class, and 0.9007 overall MMC.These (continued on next page) To develop a predictive model for identifying orders that require intervention based solely on the ordering provider's interactions with the EHR.

Database study
Data from the EHR system on provider actions and pharmacy orders In both the area under the receiver-operator (AUROC) and precision-recall (AUPR) curves, the XGBoost algorithm outperformed both logistic regressions and the random forest algorithm by a significant margin.
The area under the receiveroperator characteristic curve was 0.91, and the area under the precision-recall curve was 0.44.

Potentially inappropriate medications
Xingwei W, 2022 60  To develop a novel approach that employs machine learning models to predict the appropriateness of high alert drugs (HAD) use for a specific patient visit.

Retrospective analysis
Patient data from the Maharaj Nakorn Chiang Mai Hospital's outpatient and inpatient departments in 2018 The machine learning algorithm identified over 98% of actual HAD mismatches in the test set and 99% in the evaluation set when screening drug prescription events with a risk of HAD inappropriate use.This study demonstrates that machine learning plays an important role in screening and reducing errors in HAD prescriptions.

USA
To examine the prevalence and leading predictors of potentially inappropriate NSAIDs use among older adults with OA using realworld data from nationally representative commercial health insurance claims with the help of machine learning approaches.

Retrospective cohort study
Older adults with OA were identified using one inpatient or two outpatient claims at least 30 days apart that consisted of OA diagnosis codes (ICD-10 codes M15-M19) during the baseline year and were required that these adults be enrolled in Medicare Advantage plans with medical and pharmacy benefits during 2015 and 2016 (i.e., 24 months).
XGBoost and CVLR-both models had an AUROC value of 0.  collected data must be pre-processed.Text data, on the other hand, necessitate extensive natural language processing before use.2][103] Medical data can be collected in a variety of formats and from a variety of sources, including medical images, 3D video sequences, photographs, and numeric data.In healthcare data analysis, collecting clean, robust, and efficient data is a challenge. 104

Patient safety
Data collected from hospitals are sometimes of poor quality or inaccurate, missing data points.This leads to data error, which is one of the most difficult challenges in medical data processing using AI. 94other issue is ML algorithm decision errors, when the applied algorithm is inappropriate for the given data, or the data is not reliable enough to be used in classification algorithms such as neural networks, decision trees, and Bayesian networks. 104

Clinical implementation
The lack of empirical evidence proving the efficacy of AI-based interventions in prospective clinical trials is the first barrier to successful implementation.The majority of AI research in healthcare is generally retrospective, in a controlled environment.As a result, extrapolating results to real-world scenario is difficult.

Ethical concerns
The other main concern, apart from data privacy and security, is accountability.Poor decisions, particularly in healthcare, have serious consequences, and the current paradigm holds that someone must be held accountable. 106However, the issue of accountability becomes far more important when considering AI applications that aim to improve patient outcomes, especially when things go wrong.As a result, it is unclear who should bear responsibility if the system fails.Holding the physician accountable may appear unfair because the algorithm was not developed or controlled in any way by them, but holding the developer accountable appears too far removed from the clinical context. 94

Social concerns
One of the major social concerns is the AI in healthcare, will replace jobs, making healthcare workers obsolete.The threat of replacement leads to distrust and opposition to AI-based interventions in the healthcare.This belief, however, is largely based on a misunderstanding of AI in its various forms. 107ealthcare professionals have generally failed to keep pace with other professionals in terms of incorporating new technologies into their daily work.Previous experiences in healthcare indicate that the implementation period is an important stage in the innovation process.In practice, inventing and testing a new AI technology is not enough; other factors that can stymie its implementation in real-world healthcare, such as. 108,109) the limited data structure and quality in existing electronic health systems, (2) the alteration of the clinician-patient relationship, (3) the difficulties associated with clinical integration and interoperability, must also be considered.

Conclusion
By incorporating AI into clinical practice, health care professionals can augment their decision-making processes and provide patients with personalized care.AI allows for greater collaboration between different healthcare services provided to a single patient.For patients, AI may be a useful tool for providing guidance on how and when to take a medication, aiding in patient education, and promoting medication adherence and also AI may be used to know how and where to obtain the most cost-effective healthcare and how best to communicate with healthcare professionals, optimize the health monitoring using wearables devices, provide everyday lifestyle and health guidance, and integrate diet and exercise.
Clear guidelines on safe implementation and evaluation of AI technology in real world settings, as well as further research to understand the AI technology's capabilities and limitations, are required.While the optimal conditions for successful AI adoption are not yet in place, there is still room for further development of AI in healthcare.These include clinical validation of AI software and interventions through rigorous clinical trials, prospective observational studies to implement and understand the long-term impact of AI on clinical decisions, the development of ethical and privacy guidelines and frameworks by relevant bodies and organizations to protect patient data and promote transparency.AI can be used to develop more personalized treatment plans and patient engagement research, to improve both patients' experiences and empower them to actively participate in medication decisions involving AI.
We propose "pharmacointelligence," i.e., the integration of AI/ ML and similar advanced technologies into pharmacy practice with the sole aim of improving patient care and safety.This being said, the concepts of AI/ ML should be incorporated into the pharmacy curriculum and stakeholders should be kept abreast of innovations in this field through continuous education.As these technologies evolve at a rapid pace, the education system for pharmacists must adapt to ensure that our profession is prepared to lead these changes in care.

Table 1
Summary of findings.
(continued on next page)

Table 1
(continued ) (continued on next page)

Table 1 (
continued ) (continued on next page)