Why is patient safety important?

Patient safety is crucial in modern healthcare and refers to harm caused to patients, which could be prevented despite some unavoidable inherent risks associated with medical treatment. Healthcare professionals aim to provide the best patient care, but medical errors happen due to overworked clinicians, complex clinical cases and disorganised data. Some examples of medical errors that can be prevented are mentioned below:

• Falls in post-surgery patients
• Lack of implementation of infection control protocols cause hospital-acquired infections (HAIs)
• Administration of wrong medication or incorrect dosage
• A wrong or delayed diagnosis
• Surgical accidents, such as wrong-site surgery, wrong procedure or procedure performed on the wrong patient or leaving foreign objects, including equipment or gauze inside the body

Patient safety is important because even if medical errors do not lead to death, they prolong the hospital stay, increase the cost, prevent patients from getting back to work or lead to disabilities.¹

How is artificial intelligence (AI) improving patient experience?

AI is a technology where machines are trained to perform tasks, ideally requiring human intelligence, using data and algorithms. For example, machine learning (ML) generates algorithms, allowing computer systems to learn from data, and deep learning (DL) employs large neural networks for sequence and imaging data processing.²

AI has transformed patient care by learning from vast amounts of data. This helps improve diagnostics, create personalised treatment plans, and predict future health issues, making healthcare delivery more efficient. It has enhanced patient experience by making healthcare accessible and efficient through virtual assistants, AI-assisted remote monitoring of patients caters to the specific needs of patients, lessening waiting times and enhancing satisfaction. It provides healthcare professionals with the tools to enhance patient outcomes.³ This article discusses the healthcare areas where AI has been transforming healthcare services.

Role of AI in enhancing patient safety

Early diagnosis and diagnostic accuracy

The analysis of vast amounts of patient data using ML, such as medical records, imaging studies, laboratory results and genetic information, can contribute to early disease detection and accurate diagnosis, including various cancers. AI-assisted diagnostic tools, including medical imaging techniques, can identify subtle abnormalities that cannot be caught by the human eye, improving diagnostic accuracy, reducing errors and offering timely and appropriate intervention.

A study where researchers input a vast dataset of mammograms into an AI system for breast cancer diagnosis observed a 5.7% and 9.4% reduction in false positives and false negatives, respectively. Another study compared breast cancer diagnosis using an AI system and radiologists, with the AI system being far more sensitive (94%) in diagnosing breast cancer than radiologists (78%). AI outperformed radiologists in early breast cancer detection, with a success rate of 91% compared to 74% for radiologists.

Compared to dermatologists, DL using convolutional neural networks (CNN) could accurately diagnose melanoma. DL algorithms detected pneumonia from chest X-rays with a sensitivity and specificity of 96% and 64%, respectively, compared with the detection ability of radiologists (50% and 73%, respectively). Viz.ai analyses medical images to provide rapid and accurate diagnosis and timely intervention in emergency conditions such as stroke.^3,4

Laboratory testing

The speed, efficiency and accuracy of diagnostic tests have improved with the implementation of AI systems. Several ML models use data from various sources, including genomic data of microbes, microscopic imaging and gene and metagenomic sequencing to identify microbes, diagnose diseases and predict outcomes.

The combination of ML algorithms and digital in-line holographic microscopy could effectively detect red blood cells infected with malaria without staining, making the diagnosis cost-effective, sensitive and rapid. AI has improved laboratory efficiency in blood culture and susceptibility testing, providing results within 24-48 hours, allowing the selection of appropriate antibiotic treatment plans and increasing the recovery rates from infectious diseases.³

Risk assessment

ML algorithms can analyse electronic health records (EHRs), including medical history, lifestyle, readmission and genetic predisposition to predict the risk of diseases, adverse events or infections. This can help healthcare providers to initiate timely interventions and implement preventive measures for positive patient outcomes.³

Medication management

AI-enabled clinical decision support system (CDSS) platform guides healthcare providers in real-time. It ensures safe medication prescription by analysing drug interactions or allergies, reducing adverse events. It also eliminates dosage errors by tailoring the dosage based on individual factors, reducing the risk of over- or under-dosing.

The Merative platform by IBM uses AI to incorporate CDSS capabilities for analysing patient data, and providing guidance and recommendations to doctors. CURATE.AI optimises chemotherapy doses based on individual patient data. Cerner Millennium is a widely used medical management system that aids healthcare professionals to safely prescribe medication.^3,5,6

Surgical safety

ML-driven surgical robots can aid surgeons during complex procedures, improving precision and reducing human error. Computer vision algorithms analyse surgical videos in real-time, guiding surgeons to perform procedures with accuracy, reducing complications and improving patient outcomes.

The Senhance Surgical System comprises robotic arms, reusable instruments and 3D high-definition visualisation system to perform minimally invasive surgeries. The surgeon operates the robotic arms from a console and performs surgeries with dexterity and precision.⁵

Real-time monitoring

AI-powered monitoring devices can analyse patient data, including vital signs, EHRs and wearable devices, alerting healthcare professionals of any deviations in the patient’s condition, allowing prompt intervention and enhancing patient safety. Philips IntelliVue Guardian Solution is an intelligent and early warning monitoring system combining bedside monitors with predictive algorithms to continuously monitor patients’ vital signs and alert doctors if a patient’s condition deteriorates.⁵

Auto-charting

AI auto-charting reduces the documentation burden on clinicians as the AI system listens and completes the chart for them. This procedure streamlines and standardises data collection and reduces documentation error, improving patient safety. Deep Scribe automates clinical documentation allowing doctors to focus on patient care.⁴

Prevention of HAIs

Fuzzy logic and ML have been used for the early detection of HAIs. Majority of algorithms were created using claims-based data and laboratory and imaging results. Data from chemical vapour sensors, eNoses, was analysed using ML to rapidly identify ventilator-associated pneumonia, classifying Clostridium difficile strains and differentiate among six wound pathogens.

AI can help control infections by predicting HAI risk and employing relevant patient-specific interventions before the infection occurs.⁷

Monitoring hygiene practices

AI can contribute to adherence to established safety methods. For example, computer vision employing the convolutional network classifier monitors hand hygiene compliance in a hospital setting. Using data from various sensors, an ML algorithm delivering real-time hand hygiene alerts was designed in an out-patient setting, improving compliance rates from 54% to 100%.⁷

Venous thromboembolism (VTE)

Approximately 3.3% of inpatients develop VTEs, such as pulmonary embolism and deep vein thrombosis. AI approaches, such as the super learner ensemble approach, can be used to identify high-risk patients for future VTEs. A multiple kernel learning algorithm has been created to predict VTE risk in patients undergoing chemotherapy, which outperforms the recommended Khorana score.⁷

Pressure ulcers

Pressure ulcers are observed in approximately 2.7% of patients hospitalised in the US. The feasibility of wheelchair cushions and smart beds has been tested for detecting pressure ulcers using ML and fuzzy logic models, respectively. Using the data from embedded sensors, the algorithm detects lack of movement and the skin regions at risk of developing pressure ulcers. Although the pressure ulcer detection accuracy was 90% in the experimental setup, its utility in alerting doctors and promoting early treatment is unsure.⁷

Falls

In a hospital setting, approximately 1.1% of patients experience falls, of which, 87.5% are preventable. ML methods have been used for detecting falls early. Analysis of data obtained from a wearable sensor was able to classify patients based on risk of falls in a laboratory setting, The data from wearable sensors, small carpets and cameras used at home were used to identify fall risks and gait pattern deviations. Although these models have detected falls with an accuracy of up to 100% in an experimental setup, their use in real life requires further studies.⁷

Decompensation

Approximately 3.6% of patients develop sepsis during hospitalisation. ML has been employed to detect sepsis utilising novel gene expression markers. Improving the ML algorithms by adding data from novel biomarkers, continuous telemetry and motion activity sensors (time spent in bedroom/bathroom) will improve decompensation detection ability.⁷

Diagnostic errors

Diagnostic errors (missed or delayed diagnoses) are common and occur in 5.1% of the US population in both in-and out-patient settings. ML has been used for early detection of lung cancer using an eNose sensor that analyses exhaled breath. The support vector machine successfully distinguished between cancer patients and non-cancer controls with a sensitivity and specificity of 87% and 71%, respectively. A CDSS system based on fuzzy logic reduced delays for critical diagnoses by triaging emergency department patients with an accuracy of >99%.⁷

Benefits of AI-based healthcare

Integrating AI into healthcare enhances the efficiency and quality of healthcare services in the following ways:

• Minimises human error
• Augments the expertise of healthcare professionals instead of replacing it
• Offers 24x7 patient support as virtual assistants or remote monitoring of health condition
• Accelerates the diagnostic process and allows clinicians to focus on urgent tasks, improving patient outcomes⁴

Challenges

Although AI has tremendous potential to transform

patient care, some challenges must be overcome before successful implementation.

Ethical concerns

The privacy of patient data is crucial. AI tools require large datasets for learning and making predictions. This raises a concern about safeguarding confidential patient data against unauthorised access and breaches. These risks can be avoided by using robust encryption methods. Adhering to ethical guidelines and ensuring transparency are crucial to maintaining patient’s trust and avoiding AI’s adverse effects.

Staff training

As the role of AI increases in the healthcare setting, hospital staff must get adapted to working together with these advanced technologies. To improve patient care, training staff to use AI tools and encouraging collaboration between AI and humans is crucial for flawless integration with existing healthcare systems.

Regulatory compliance

Implementing AI in healthcare requires compliance with regulations, laws and guidelines. Collaboration among policymakers, technologists and legal experts is crucial to maintaining patient trust, promoting innovation and avoiding financial or legal consequences.

Lack of quality medical data

Fragmentation of medical data across different institutions and platforms prevents comprehensive data collection. For AI systems to be effectively trained and validated, medical data formats need to be standardised and operable in various systems.⁴

FAQs

Is AI currently being used in any governmental health care systems? 

Yes, AI is being integrated into various governmental healthcare systems worldwide. In the U.S., AI supports predictive analytics and diagnostic tools for veterans, while the UK's NHS uses AI for imaging analysis and administrative efficiency. Canada and Australia are also employing AI to improve patient outcomes, optimise hospital operations, and support telehealth services.

When will AI be officially incorporated in hospitals and clinics?

While AI is set to make significant advancements in healthcare, its widespread official adoption will hinge on overcoming existing challenges and establishing robust regulatory frameworks to ensure patient safety and data security.

What are some risks related to patient data when dealing with AI?

A: When dealing with AI in healthcare, risks related to patient data include privacy breaches, data security threats, and the potential for bias and discrimination from biased datasets. 

Summary

In the modern healthcare system, AI demonstrates immense potential to revolutionise patient safety. The main areas for patient harm include diagnostic errors, adverse drug events and decompensation. Building robust AI models using traditional data, such as EHRs and claims and novel data, such as wearable devices and sensors, is essential. AI can assist clinicians in accurate diagnosis, recommend treatment plans and identify high-risk patients using predictive analytics.

The risks associated with AI, including data quality, privacy and bias, must be overcome by careful validation and continuous quality check of the models and responsible implementation to ensure patient confidentiality and safety.