Artificial Intelligence at the Bedside

Dr Muhammad Shakir Balogun

31/05/2025

Introduction: What Is AI in Clinical Care?

2

Defining AI

• Computer systems that perform tasks usually requiring human intelligence.
• Learning from data, recognizing patterns, and making decisions.
• Used to analyze clinical data and assist doctors in diagnosis and patient care.
• Mimics certain cognitive functions but does not “think” or feel like a human.
• Can rapidly sift through large amounts of information to spot patterns.

3

AI vs. human intelligence—augmented, not replacing

• Currently, AI in medicine is "narrow AI" – specialized to perform defined tasks
• AI is not a human-level doctor robot—augments, rather than replaces, human doctors.
• The American MA prefers "augmented intelligence" to emphasize AI's role in supporting, not replacing, clinicians.
• AI systems lack a doctor's intuition, common sense, and reasoning abilities.
• Excels at analyzing data and finding patterns—does not "reason" like a physician.

4

Why AI now?

• Improvements in computing power, big data from electronic health records, and new algorithms like machine and deep learning.
• AI systems can be trained on vast amounts of clinical data to recognize patterns.
• As healthcare becomes more digital, AI has more data to learn from.
• AI enables more sophisticated applications in patient care, enhancing diagnostic and treatment capabilities.

5

Current Use Cases of AI at the Bedside (Global & Nigerian)

6

AI for diagnostics and imaging

• One of the most developed uses of AI in bedside care is interpreting medical images and signals.
• AI algorithms analyze radiology scans and pathology slides for abnormalities with impressive accuracy.
• AI helps flag critical findings, allowing specialists to prioritize urgent cases, especially in areas with scarce specialists.
• Zebra Medical Vision: An Israeli company providing AI tools to detect diseases from imaging:
• Reached hospitals in Asia and Africa, including Nigeria, helping to bridge radiologist shortages by acting as a second reader of scans.

7

AI for clinical decision support & early warning

• AI is integrated into bedside monitoring systems and EHRs to provide early warning scores and decision support alerts.
• AI can track vital signs, lab results, and nursing notes to predict patient deterioration hours in advance.
• Advanced hospitals use AI-driven systems that analyze trends in real time and alert staff to subtle signs of trouble (rising HR, dropping BP).
• AI tools help improve patient outcomes by prompting earlier interventions based on detected patterns.
• CLEW platform in ICUs uses machine learning to predict patient instability and sends alerts to clinicians for timely action.

8

AI for triage and symptom checking

• AI helps triage patients by determining who needs urgent care and who can wait or self-care.
• Apps like Ada Health’s app are used globally (including in Nigeria) to guide patients on whether they need urgent care.
• They ask about symptoms and medical history, then use AI to suggest possible causes or recommend if a doctor visit is necessary.
• Ada Health’s AI-driven chatbot provides preliminary medical guidance and improves diagnostic triage, especially in resource-limited settings.
• Triage nurses can use AI tools to prioritize patients by analyzing symptoms and vital signs to predict conditions like dehydration or severe illness.

9

Global example – early warning in obstetrics

• SPEC-AI trial: The first RCT of an AI tool in Nigerian healthcare.
• Used to detect peripartum cardiomyopathy, a dangerous heart condition in pregnant women, at the bedside.
• The AI analyzed echocardiographic or clinical data to flag high-risk women.
• AI screening doubled the detection rate of peripartum cardiomyopathy among obstetric patients.
• More women received early diagnosis and care who might have been missed by routine practice.
• The success of this trial highlights how AI can support frontline clinicians in real-world Nigerian settings, from maternity wards to general hospitals.

10

Nigerian use cases and pilots

• Ubenwa Health is a Nigerian startup using AI to analyze an infant’s cry as a diagnostic tool.
• Detects biomarkers in crying patterns to screen for birth asphyxia.
• Being piloted to assist health workers in maternity wards and at home.
• The AI listens to a baby’s cry and alerts if distress/asphyxia is detected.
• A team of Nigerian schoolgirls developed an AI-driven app to identify counterfeit drugs by scanning drug barcodes.
• Addresses a significant patient safety issue.
• Nigerian hospitals are experimenting with AI-powered telemedicine bots.
• AI chatbots on WhatsApp or web handle initial patient queries and scheduling.

11

Future Possibilities of AI at the Bedside

12

Predictive modeling for personalized care

• AI has the potential to enable predictive and personalized medicine at the bedside.
• AI could forecast a patient’s health progression and tailor interventions based on individual needs.
• AI could analyze large personal data sets to predict potential complications before they appear.
• The concept of an AI “digital twin” for patients could become a reality.
• Long-term trajectory: AI will help clinicians anticipate and prevent problems, moving beyond reactive care.

13

Real-time patient monitoring and “smart” wearables

• The patient’s bedside could extend beyond hospital walls with wearable sensors and IoT devices streaming data to AI systems.
• Patients may wear smart monitors (e.g., smartwatches, patches) that provide real-time data on heart rate, oxygen levels, ECG, and glucose.
• Overhead cameras or contactless sensors could monitor patient movement, facial expression, or breathing rate to detect potential risks.
• At home, wearables could send alerts to a doctor’s dashboard if signs like weight gain or pulse changes indicate impending issues (e.g., heart failure).
• AI will provide 24/7 observation, expanding the concept of "bedside" care to wherever the patient is, linking patients and providers in real-time.

14

Integration with EMR and workflow

15

Beyond diagnosis – AI in therapy and robotics

• In the future, AI could assist in treatment delivery at the bedside.
• AI-driven pumps could adjust drip rates based on real-time patient feedback.
• AI-powered robots could assist in tasks like turning patients, delivering medications, or supporting surgeries.
• AI may play a role in discovering or optimizing medications, which could lead to more personalized drug therapy.
• AI could power virtual assistants on patients' phones or bedside tablets:
• Providing education, answering questions, or conducting mental health check-ins through natural conversation.

16

Practical Implementation Challenges in Nigeria

17

Infrastructure limitations

• Many facilities struggle with basics like consistent electricity, internet connectivity, and modern IT equipment.
• AI tools often need reliable power and network access (for cloud-based systems or updates) and sometimes high-performance computers.
• Clinics with insufficient computers or frequent power cuts find it difficult to implement AI decision support systems.
• AI relies heavily on data, but if patient records are still paper-based or not digitized, integrating AI becomes difficult.
• These challenges slow the adoption of high-tech innovations like AI in healthcare.

18

Data quantity and quality

• Shortage of large, curated medical datasets from the local population, particularly for diseases, outcomes, and imaging.
• Inconsistent record-keeping, entry errors, or biased sampling can degrade AI performance.
• AI models may exhibit biases due to unrepresentative data, affecting their performance in specific populations.
• Global AI models may not prioritize diseases prevalent in Nigeria, like malaria or sickle-cell anemia.
• Gathering relevant, high-quality local data is a critical challenge for AI adoption.

19

Workforce and training gaps

20

Cost and sustainability

21

Resistance to technology adoption

• Cultural and institutional resistance can hinder the integration of AI in healthcare.
• Some healthcare workers may fear that AI will replace their roles or undermine their expertise.
• Doctors may question whether a computer can truly guide their treatment decisions, asking, “Can AI really tell me how to treat my patient?”
• Patients may be concerned about AI, especially if not well-explained, fearing they might be treated like a “guinea pig” or miss out on human care.
• Health system leaders may hesitate to approve AI projects due to unfamiliarity or fear of failure.
• Acceptance and trust from both healthcare professionals and patients are a major hurdle for AI adoption.

22

Ethical and Regulatory Considerations

23

Data privacy and patient consent

• Health data is sensitive, raising questions about how patient information is collected, stored, and used in AI systems at the bedside.
• The Nigeria Data Protection Regulation 2019 provides guidelines, ensuring consent and proper data handling.
• Patient data often needs to be de-identified before being used to train AI.
• Patients should be informed when AI is involved and provide their consent.
• AI systems could be vulnerable to hacks or data breaches, making robust cybersecurity essential for ethical AI deployment.
• Many AI tools send data to cloud servers for processing, which can violate local privacy rules if not properly managed.
• Ensuring patient confidentiality while using AI is crucial, especially when sensitive data is involved.

24

Algorithmic bias and fairness

25

Accountability and liability

• When AI makes a recommendation that harms a patient, who is responsible (doctor, hospital, or software developer?
• Nigerian legal frameworks have not yet defined liability in AI-assisted healthcare.
• Where AI autonomously interprets results, such as missing a cancer diagnosis, it’s unclear whether liability lies with the radiologist for trusting AI or with the AI company for product failure.
• Liability issues are being discussed worldwide, and Nigeria faces similar challenges.
• Nigeria lacks a clear regulatory body or standards for AI in healthcare.
• Clinicians may either over-rely on AI without accountability or avoid using it due to fear of liability.
• Additionally, AI developers should be held to certain accuracy and safety benchmarks.

26

Regulatory gaps in Nigeria’s health sector

27

Ethical use and “do no harm”

• AI should ultimately benefit patients and do no harm.
• Patients should be informed when AI is involved in their care.
• Clinicians should understand why the AI suggested a particular decision or action, avoiding "black box" systems that lack clear reasoning.
• AI should not be used to deny care.
• AI should support, not replace, human judgment.
• Clinicians must maintain empathy and human elements in AI-assisted care and not reduce patients to data points.
• AI tools should be accessible to all, including persons with disabilities
• WHO guidance on Ethics & Governance of Artificial Intelligence for Health (2024) outlines key ethical principles for AI in health.
• Nigeria should strive to integrate these ethical principles as it adopts AI in healthcare.

28

Conclusion: Engaging with AI Safely and Productively

29

Embrace continuous learning about AI

30

Start small

• Start small – pilot AI tools in a controlled way: Healthcare institutions should adopt AI through pilot projects to manage risk and assess effectiveness.
• Set clear goals and metrics: Define specific objectives and measure progress over a few months.
• Evaluate results: Assess the pilot’s effectiveness before scaling it up to ensure AI integration is effective and meets goals.
• Work out kinks on a manageable scale: Use the pilot to address potential issues such as technical difficulties, staff training requirements, and integration challenges.
• Involve end-users: Engage doctors, nurses, and other healthcare staff in the pilot, collecting their feedback and securing their buy-in.
• Make the case for broader implementation: If the pilot shows positive outcomes or increased efficiency, use the results to advocate for adoption.

31

Invest in infrastructure and data readiness

• Advocate for prioritizing the digital foundation required for AI in healthcare.
• Consistently entering diagnoses and outcomes into electronic systems helps prepare data for AI models.
• Consider establishing a dedicated data team or partnering with tech companies to organize and digitize historical data.
• Highlight that investments in infrastructure and data readiness benefit both current care and future AI applications.
• Improved data management will enhance healthcare today and pave the way for the adoption of advanced AI tools.

32

Develop and follow guidelines for safe AI use

33

Engage with local tech and research communities

34

Keep the patient at the center

35

36