ED Data Project - Pakistan

Data is the modern world’s most valuable resource, however, ineffective collection techniques often lead to an accumulation of junk rather than data. The Indus Health Network, like many other hospital systems in developing countries, uses an EMR system collecting patient data in an unusable open script format. Billions and trillions of existing data points currently require manual manipulation and cleaning, an impossible and tedious task preventing integration with systems like FHIR. Our team analysed the problem and we propose a 2 step solution: 1) develop an algorithm to read patient records, apply ICD codes and transfer coded data into a big data registry. 2)Use big data to predict patient outcomes from similar cases that presented before and analyse human resource and facility allocations. Our registry is focused on Emergency services but can be customised according to any department. This solution maximises the ER's potential as a public health surveillance system.

In countries lacking primary healthcare services like Pakistan, along with critical emergencies, patients with minor ailments present directly to ED’s expecting immediate treatment. ED’s are over-burdened, understaffed and carry great responsibility as surveillance systems, a fact highlighted by the current COVID-19 pandemic. Unfortunately due to the shortage of specialised doctors, it will take the training programs of Pakistan more than 5 decades to train the required number of ED doctors for a 220 million population, according to the suggested physician-patient ratios (NCQA USA). Added to this the lack of meaningful clinical data paint an unfortunate and dismal picture of the healthcare system, unequipped to identify public health threats. Overcrowding and lack of comprehensive health policies leads to questionable resource allocations and ED delays due to physical distance and deficient inter departmental communication. The solution we are proposing will address planning, daily functioning and capacity of emergency departments along with serving as a clinical decision making support system. Young doctors, particularly in far-flung areas, can view the interventions applied for positive outcomes of similar complaints through an HMIS integrated smart system, displaying recommendations based on records of millions of patients analysed by deep learning.

The complete ICD coding system will be uploaded to a neural network through an AI development software and complete patient records from the adult and pediatric ER will be entered in gradual phases. Manually coded patient records will be compared to the auto-coded records, tracking errors as the algorithm improves. The complaints, comorbidities, diagnosis, investigations done and treatment of the patient will be read and matched to the corresponding codes. In order to analyse outcomes in different resource settings, data from the learning site (main campus) will be added to with data from other campuses (validation sites) offering varying demographics and management decisions. Once coding accuracy exceeds 85%, the data will entered into a registry containing patient, environmental and facility related variables. This registry will then be useful for analysing patient outcomes according to days, available staff, investigation times, public holidays and other varying factors including patient history and  treatment related outcomes. Along with deep learning of facility and resource allocations, this coded data in large volumes can potentially offer clinical decision making support to healthcare workers especially those lacking specialists. This solution identifies the need for AI applications to improve patient outcomes using prior examples within the institution.

The target population of this project can be categorised into three groups:

1)    Healthcare workers, in particular medical officers who are not trained for specialised care but are unfortunately delegated the task of independently managing emergency cases. They require a smart clinical decision-making support system that is easy to use and contains support based on local data and management – as majority of the support tools contain protocols and management options from resource-rich developed countries that are often not applicable in an emergency room in a low-income area of Pakistan.

2)    Policy-makers and healthcare workers in other countries and settings requiring similar support for identifying local problems and  improving patient care in the emergency setting can adopt the algorithm and use their own data for deep learning, particularly in situations where healthcare systems are buried under large patient influx and sub-par medical record keeping. We aim to create a step-by-step execution process leading to eventual long-term improvement in health policy and management that can be adopted and used wherever needed.

3)    The greatest target of this solution is the general population, in particular the large underprivileged community existing in developing countries who are the most frequent visitors of low-resource setting hospitals and public institutes with compromised standards of care. The Indus Health Network is a free of cost hospital built and operating exclusively for this sector of society.

The idea of this system has arisen from multiple studies published by myself and my team over the years, along with multiple articles by foreign authors and observers on the lack of clinical data usage for local improvement in our setting. We also engaged and refined this idea through discussion with various AI companies who have prior experience in creating AI solutions for healthcare issues.

Patient recovery and prognosis depends acutely on the facility where the patient presents. We propose to use patient data to understand gaps and shortcomings in the system rather than creating policies mimicking past policies that have so far fallen short. Number of training programs, hospital designs and government support are not controllable factors so we aim to make use of all the tools that are within our reach for viable and holistic solutions.

The challenge stated strengthening disease surveillance and early warning predictive systems as one of its healthcare foci. An unfortunate reality is that lack of effective documentation and understanding of local systems in LMIC’s can lead to catastrophic global effects, that is being witnessed in India during the COVID pandemic. Health and humanity are intrinsically connected; disease outbreaks, antibiotic resistance, non-compliance to vaccinations are just some of the ways in which individuals from one country can potentially disrupt the global health balance. Evidence based practice is essential to every country, the mass effect of which can lead to global prosperity

This project is a long term commitment requiring additional human resources for data management along with the technical aspect of software development. Despite stringent efforts on our part, since Indus Health Network is a non-profit organisation, a lack of funding has prevented us from moving past the Concept phase of this project so far. AI applications to healthcare are not only uncommon but have so far not been practically done in Pakistan. Finding local data scientists who can work in collaboration with our medical and public health experts has been challenging leading to our connections with international companies and AI experts, endeavours that we have been pursuing in the limited scope that we can.

Although health tech and AI applications in healthcare are not unheard of, the focus of health tech tends to be towards clinical applications and outpatient services. Hospitals and the healthcare system are not only services but also data banks, containing valuable reflections of population trends and public health problems. This data becomes even more valuable in countries where authorities are unable to enforce regulations and govern effectively, often compromising on public health programs and preventive medicine. The clinical data currently being used from hospitals in Pakistan , among other LMIC's, is identified according to the interest of the researcher with individual glory being the ultimate goal rather than overall community, provincial or national progress. In settings where this is not the case, effective large scale data is still only partially usable due to the lack of regulation in the way data is documented and stored. Transitions to Electronic Health Record systems do not reflect an improvement in medical record keeping as each institute considers their own economic convenience and time management rather than considering the importance of national data overall. Our solution aims to turn this entire situation on its head. An algorithm that automatically creates registries of readable and organised data out of junk may not be considered a unique or novel idea - however, its impact can be revolutionary and hugely impactful in developing countries.

The Emergency Department of Indus Health Network caters to up to 300,000 people annually, a number expected to rise exponentially as we move into a new campus with ten times more capacity. As the second phase of this solution will initiate clinical support tools, better functionality will lead to shorter ED delays and consequent improvement in capacity to serve more patients.

Analysing patient outcomes and ED delays and supplementing it with better quality care will lead to achievement of good health and well-being, an SDG set by the United Nations.

Impact and progress toward this goal will be measured by impact on patient numbers and outcomes in the Emergency Department.

Full time: 3 members

Part time: 2 members

Contractors: 1 AI organisation volunteering assistance for this project

A larger full time and part time team is needed for this solution including IT personnel, data operators and project managers, these will be hired once we have a dedicated budget for this initiative.

My team contains doctors and public health professionals with vast experience in execution of healthcare delivery and projects outreaching to large communities. The team also contains experts in AI and healthcare technology with a deep connection with the HMIS system functioning locally.

Each of the team's members has spent most of their professional lives working in the healthcare system in Pakistan and travelled around most of the country, exposed to various harsh conditions in which emergency services often exist.

Each phase of this idea so far has been developed through diligent research, fruitful discussions with skilled people bearing diverse experiences and consensus between the most practical path by all stakeholders.

Our team contains equal members from all genders, races and ethnicities. Selection was based on voluntary participation along with the expertise offered by each individual regardless of any personal characteristic.

The Indus Health Network as an institute is exemplary in offering equal opportunity free from religious and cultural discrimination.