AMPATH EHR Data: Current & Future Capabilities

AMPATH EHR Data: Current & Future Capabilities

This post the text and slides for Part 2 of the lecture (full video below). Part one is available here.

Part 1: History of AMRS [text and slides here]
- Problem of paper records 2:11
- Microsoft Access & OpenMRS System 8:48
- Building of AMRS POC 13:24
- The AMRS Team Today 17:31
- Demo of AMRS POC 20:58
- Upcoming Major Projects 30:33

Part 2: Current Data Capabilities & Future of AMRS
- Current Clinical Data Capabilities 34:57
- New data visualization tools: Kibana 41:58
- Future of AMRS  50:52

Presented at AMPATH Partners Summit October 10, 2018 Eldoret, Kenya.

Part 2: Data Analytics

1. Current Clinical Data Capabilities

The system is also able to produce reports to meet the Ministry of Health standards - and this has replaced dozens of those large registers we saw earlier. Everytime we go into a new department, some of the first questions asked are “What reports are being produced for the Ministry of Health and other funders/agencies”. A high priority is being able to reproduce government and grant level reports with the highest accuracy and speed possible.

The dashboards are also used to clinic workflow and processes. Any clinic can see their data. Admittedly, we still have work to do to improve the interface and help train users better how to access this.

First is the clinic daily schedule, that can be filtered by location, and show the clinic flow, wait times, and when patients are arriving. The tab ‘Has not returned’ shows those patients that need to be contacted by the retention workers.

Monthly schedule. The Orange button shows the patients who did not show that day. Again, making it very easy to contact and follow up with these people

Active patient program enrolment board. Again, can be filtered to great detail. if the user clicks any of the numbers, it will generate a patient list for that item.

Clinic Overview Visualization. Compares key clinic indicators over time.

Patient Care Status, further key indicators over time.

Defaulter List. A key patient list for clinic workflow.

The MOH-731 report - a government level data report.

HIV Summary Indicators. This is a flexible sandbox module. Any clinic or county and develop their own questions to answer by searching for those indicators. In this example, the user compares the number of patients on ARV and those who are pregnant on ARV.

When searching for this, there is a ‘wait’ and loading page. To generate any of the reports earlier there is often a delay as the system runs the search. We will see later a new tool, Kibana, that can generate reports in an instant.

Patients Requiring Viral Load Order

Patient Referral. This has been added in last month. It allows clinics to see who has been referred to their clinic, and this can be sorted by the different programs they offer.

2. New data tools: Kibana

Part of the trouble we run into in the current system is the slow speed of searching our databases, as they are in mySQL tables. Therefore a significant project over this last year was transforming our existing MySQL databases into Elasticsearch. 

This allows us to users to visualize this information via Kibana - which is a very fast big data visualization tool.

A few months ago AMRS started to develop Kibana dashboards. This is a summary of MOH-731 which is a standard Ministry of Health Report. This example shows the AMRS data is up to government level reporting quality. This report can be filtered instantly by any county and by any clinic. 

When Kibana is implemented across the network, every clinic clinics can access their data in real time, and generate their own search queries and visualizations.

Below is some examples of how targets can be set, for instance - to track viral load.

Clinic outcomes can also be compared in real time.

This last set of dashboards show real time data collected from the day before of the AMPATH HIV network.

The first display is of a map of all locations and the clinic volume that day.

Second we can look at the time of arrival, and time they are seen by the clinician.

This graphic shows the time to complete the clinical encounter at each site, broken down to the time they waited for triage, and the time for the clinical appointment.

The next two graphics show the top clinics and the number of patients seen that day. The graphic on the right shows the top programs that had patient encounters.

The chart below shows the time to complete each clinical form. In theory one could run two different forms, and do an A/B test to see which one is quicker to complete.

All this information can be filtered in an instant by date, county, clinic, or any other attribute. For instance, if we wanted to sort the HIV patients who also had breast cancer screening, we can sort that.

As you can see, AMPATH is truly able to build a health system with this type of data on system performance.

Access to Data

A key focus in the coming year is to ensure that everyone from patients, to providers, to clinics, to counties, have access to data.

Patients find it encouraging to see visually how their viral load is dropping. Providers & clinics need to be easily able to see how their patients are doing, identify those patients who have been lost to follow up. Counties need to be able to know how each site is doing.

On a global level, we are looking to create data visualizations to share in real time on a public link online.

3. Future of AMRS

Machine Learning

You may have heard discussed in the news the idea of ‘machine learning’. This is the ability for computers to teach themselves specific skills. Machine learning has lead to advancements in voice to text on your phone such as Siri to self driving cars.

There is much hope that machine learning will help assist in medical care. This may have the most significant benefit in resource limited settings. Because ultimately if machines can provide routine diagnosis and management it will drop the cost to provide this care essentially to that of a moving electrons, or a Google Search, which is pretty much free and universality accessible.

AMPATH has an extensive and high quality dataset which is necessary for work in machine learning. We are working with researchers at NIH, Regenstrief Institute, Brown University, UCSF, on ways that machine learning can assist our clinicians and patients here.

Two examples include a project to look at the ability of the computer to predict HIV Viral Loads. This could help inform clinicians the optimal time to repeat tests. Another with machine learning to aid in the interpretation of chest x-rays, and the clinical decision support to act upon these results.

AMRS 3.0

As AMAPTH has grown from taking care of patients only with HIV, into providing more comprehensive care with population health, the needs of the medical record system have also expanded.

This period of growth is the perfect time to step back and consider, “how we build the perfect EHR, one that is able to take better integrated care of the patient”

We have started to call this vision - AMRS3.0. This is a whole separate presentation on how we have derived from first principles of how medical data is stored and how clinicians think about information to create a really amazing user interface that works for everyone from community health workers to consultants.  You can read or watch that presentation at: AMRS Design presentation here: Foundational EHR Design Principles.

The take away is that we have approached this with the ability to integrate care. Meaning that chart is designed to manage multiple issues at once.

This is a huge undertaking, and therefore we are in conversation with the OpenMRS community on how we can coordinate our global efforts as we work towards trying to provide comprehensive health care.


I would like to thank the many contributors over the years who have built the AMPATH medical record system. Our current team continues to push the envelope in ways to expand and improve the system, to act better on our data, and ultimately provide the best tools possible to clinicians so that patients can heal and flourish.

The magic of the approach at AMPATH is the tight integration of clinical service delivery and electronic medical record development into the same organization. This allows us to move quicker and create very tight integrations in innovations in care delivery supported by IT systems, and data analytics.

Lastly, I would like to thank all of our AMPATH partners here. None of this would have been possible without them.

Summary of today: 

1.    The problems of paper records for individuals and groups

2.    How OpenMRS was developed first in Eldoret and now is in the globe

3.    How in the last three years AMPATH has implemented a point of care real time system for 60,000 patients, with an additional 50,000 on retrospective data entry (being converted to point of care this year).

4.    How we we anticipate growing the number of sites the system is in by five fold in the near future to serve the needs of a comprehensive care system.

5.    How AMPATH works to deliver a system of integrated to care for the patient

6.    AMRS is supported by a diverse team of experts

7.    The powerful data the AMPATH medical record system is able to provide, both for real time clinical use, as well as evaluation and metrics, and how this data can be accessed by users at all sites.

8.    A look at the future, where we re-think the medical record system from first principles to make it faster, simpler, more reliable, even more power; as well as wonder how we can harness the benefits of machine learning to help make care more accessible, higher quality, and lower cost.

Consider reading the first part of this lecture, Part 1: History of AMPATH Medical Record System,

HHDS.17 - Markov Decision Processes and Its Applications in Healthcare

History of AMPATH Medical Record System (AMRS)

History of AMPATH Medical Record System (AMRS)