Improving Patient Care Through A.I. and Machine Learning

Improving Patient Care Through A.I. and Machine Learning

Clinical Deterioration Dashboard

Company

Robô Laura

Platform

Live Dashboards

My Role

UX Researcher

-

UI Designer

*Some information was hidden to stay aligned with GDPT.

Overview

The Clinical Deterioration Dashboard is a platform to monitor UCI patients. The system uses A.I. and Machine Learning to analyze the patient’s vital signals and generate alerts for the medical team when a patient fits a risk profile for sepsis.


The dashboard was on the market for a long period without significant improvements. Despite the good feedback about the dashboard, the active users were decreasing and the support tickets were increasing, which led the company to focus the effort on understanding users' needs to improve engagement and retention.

The Clinical Deterioration Dashboard is a platform to monitor UCI patients. The system uses A.I. and Machine Learning to analyze the patient’s vital signals and generate alerts for the medical team when a patient fits a risk profile for sepsis.


The dashboard was on the market for a long period without significant improvements. Despite the good feedback about the dashboard, the active users were decreasing and the support tickets were increasing, which led the company to focus the effort on understanding users' needs to improve engagement and retention.

My Role

My responsibilities in the project were: to organize and manage the design process end to end, conduct and support interviews, conduct workshops, develop wireframes and high-fidelity prototypes, conduct usability tests, iterate on the solution, and create the documentation to handoff to the development team.

My responsibilities in the project were: to organize and manage the design process end to end, conduct and support interviews, conduct workshops, develop wireframes and high-fidelity prototypes, conduct usability tests, iterate on the solution, and create the documentation to handoff to the development team.

Problems

The client’s needs were evolving and the features weren’t. Some of the issues the users were having are:

The client’s needs were evolving and the features weren’t. Some of the issues the users were having are:

Lack of information

The users didn’t understand what to do after the system indicated a patient at risk. The dashboard only showed the patients’ bed number, when it occurred, and icons that didn’t clarify their purpose.

The users didn’t understand what to do after the system indicated a patient at risk. The dashboard only showed the patients’ bed number, when it occurred, and icons that didn’t clarify their purpose.

Impact

Patient care was being affected since the information was not enough to give the medical team inputs to understand how to proceed.

Patient care was being affected since the information was not enough to give the medical team inputs to understand how to proceed.

Last version before the redesign.

Mistrust with the solution

They believe the system was not working well, since after the alert the patient's vital signs were good and the patient wasn’t showing any signs of deterioration.

They believe the system was not working well, since after the alert the patient's vital signs were good and the patient wasn’t showing any signs of deterioration.

Impact

The patient's condition was being neglected because of his present condition. Plus, the medical team was starting to mistrust the solution.

The patient's condition was being neglected because of his present condition. Plus, the medical team was starting to mistrust the solution.

Bugs

The system was having a lot of problems and inconsistencies, which led to an increase in technical support requests.

The system was having a lot of problems and inconsistencies, which led to an increase in technical support requests.

Impact

In some cases, the dashboard was showing a lot of patients and that wasn’t reflecting the real state of the patient's condition.

In some cases, the dashboard was showing a lot of patients and that wasn’t reflecting the real state of the patient's condition.

Opportunity

How could we enhance the medical team's awareness to help them provide fast and efficient patient care?

How could we enhance the medical team's awareness to help them provide fast and efficient patient care?

Business Goals

Scalability to expand value proposition

Scale the solution for other protocols beyond sepsis to expand the value proposition and create flexibility for the customers.

Scale the solution for other protocols beyond sepsis to expand the value proposition and create flexibility for the customers.

User-centered design

Working closely with the users to iterate on solutions and offer constant improvements aiming to empower the medical team.

Working closely with the users to iterate on solutions and offer constant improvements aiming to empower the medical team.

Users Goals

Prevent patient deterioration

Starting patient care at any signal of deterioration to avoid complications.

Starting patient care at any signal of deterioration to avoid complications.

Choose the best medical conduct

According to the patient's condition and severity, the medical team must choose the best path for treatment.

According to the patient's condition and severity, the medical team must choose the best path for treatment.

Discovery

Stakeholders

Analysis

Stakeholders Analysis

Qualitative Research

Service

Blueprint

Service Blueprint

Stakeholders’ Analysis

I mapped three key stakeholders that could bring important information and insights:

I mapped three key stakeholders that could bring important information and insights:

Nurses

They are responsible for inputting patient vital signals into the system, managing the dashboard, doing the initial care when a patient pops up on the screen, informing the doctor about everything, and proceeding accordingly to the doctor’s directions.

They are responsible for inputting patient vital signals into the system, managing the dashboard, doing the initial care when a patient pops up on the screen, informing the doctor about everything, and proceeding accordingly to the doctor’s directions.

Doctors

They are responsible for deeply investigating the patient's condition, choosing medical conduct, and giving instructions to the nurses to continue the procedures.

They are responsible for deeply investigating the patient's condition, choosing medical conduct, and giving instructions to the nurses to continue the procedures.

Internal Customer Support Team

Over the years the Customer Support Team had collected several feedback and insights from customers and was still working closely with them.

Over the years the Customer Support Team had collected several feedback and insights from customers and was still working closely with them.

Qualitative Research

Since the company was looking for close relationships with its customers, I found it valuable to conduct qualitative research and get feedback directly from the nurses and doctors, including medical directors. Some problems I found during the interviews with external and internal stakeholders are:

Since the company was looking for close relationships with its customers, I found it valuable to conduct qualitative research and get feedback directly from the nurses and doctors, including medical directors. Some problems I found during the interviews with external and internal stakeholders are:

Protocol Limitation

The hospitals have several protocols for different situations and the system was limited to only one protocol.

The hospitals have several protocols for different situations and the system was limited to only one protocol.

Impact

Nonuse of the system and poor retention. In case the hospital had requested the inclusion of other protocols, the development team would have to do it manually.

Nonuse of the system and poor retention. In case the hospital had requested the inclusion of other protocols, the development team would have to do it manually.

Knowledge Gap

A large number of users didn’t know how the alerts were generated.

A large number of users didn’t know how the alerts were generated.

Impact

Lack of engagement and wrong decision-making. 

Lack of engagement and wrong decision-making. 

Space limitation

The number of alerts the platform displayed was limited.

The number of alerts the platform displayed was limited.

Impact

Patient alerts were out of the dashboard and had to wait until the first ones were finished. 

Patient alerts were out of the dashboard and had to wait until the first ones were finished. 

Lack of instrunction

It wasn’t clear to the nurses how was the procedure to remove a patient’s alert that had already been seen from the dashboard.

It wasn’t clear to the nurses how was the procedure to remove a patient’s alert that had already been seen from the dashboard.

Impact

A patient alert could stay for hours in the dashboard and if it was a high severity alert it could remain at the top, taking space from other alerts to be visualized.

A patient alert could stay for hours in the dashboard and if it was a high severity alert it could remain at the top, taking space from other alerts to be visualized.

Service Blueprint

During the interviews I created a blueprint to look for other issues the users might be facing, and by the end, I came up with four main opportunities to focus the effort on the ideation step:

During the interviews I created a blueprint to look for other issues the users might be facing, and by the end, I came up with four main opportunities to focus the effort on the ideation step:

Accessible Information

Incorporating pertinent information to assist nurses in identifying the cause of the patient's alertness.

Incorporating pertinent information to assist nurses in identifying the cause of the patient's alertness.

Customization

Enabling hospitals to customize platform usage based on their preferences.

Enabling hospitals to customize platform usage based on their preferences.

Scalability

Ensure the scalability of the product and reduce the reliance on opening tickets for technical support.

Ensure the scalability of the product and reduce the reliance on opening tickets for technical support.

Dashboard Management

Empowering users to manage patient alerts.

Empowering users to manage patient alerts.

Ideation

Insight

Workshop

Insight Workshop

Wireframe

Development

Wireframe Development

Design

Critique

Design Critique

Usability

Test

Usability Test

Internal Workshop for Insights

I conducted a workshop to collect insights, business rules, and technical requirements and limitations. There was a large number of internal professionals actively involved in the project such as developers, nurses, doctors, a customer success team, machine learning, and A.I. specialists. This collaboration process brought valuable insights to the project.

I conducted a workshop to collect insights, business rules, and technical requirements and limitations. There was a large number of internal professionals actively involved in the project such as developers, nurses, doctors, a customer success team, machine learning, and A.I. specialists. This collaboration process brought valuable insights to the project.

Wireframe Development

For the initial drafts, I focused on organizing the information on the dashboard to give the internal team an overview of what was important and what wasn’t. I knew that overloading the medical team could be as bad as give them no information at all.


The use of no colors to highlight the severity was one of the hypotheses that it could confuse the medical team since they use colors to categorize another process at the hospital.


These are the initial wireframes:

For the initial drafts, I focused on organizing the information on the dashboard to give the internal team an overview of what was important and what wasn’t. I knew that overloading the medical team could be as bad as give them no information at all.


The use of no colors to highlight the severity was one of the hypotheses that it could confuse the medical team since they use colors to categorize another process at the hospital.


These are the initial wireframes:

Wireframe used for the firsts usability tests.

Other drafts with some variations.

Design Critique

Aiming to expand our options and find the best solution for our users, I’ve conducted a design critique session with our design team. The collaboration was very important and generated lots of insights.

Aiming to expand our options and find the best solution for our users, I’ve conducted a design critique session with our design team. The collaboration was very important and generated lots of insights.

Usability Test and Iteration

For the usability tests we targeted hospitals that we considered heavy users and I conducted both remote and face-to-face usability tests. The users were very excited with the new design and features, and came up with constructive feedback and suggestions to improve the dashboard.

For the usability tests we targeted hospitals that we considered heavy users and I conducted both remote and face-to-face usability tests. The users were very excited with the new design and features, and came up with constructive feedback and suggestions to improve the dashboard.

High-Fidelity Prototype

After some iterations and adjustments, I’ve finished the high-fidelity prototype. Here is the new design:

After some iterations and adjustments, I’ve finished the high-fidelity prototype. Here is the new design:

Score and Historical Chart

The inclusion of a value between 0 and 10 and colors to indicate the severity.

The inclusion of a value between 0 and 10 and colors to indicate the severity.

Impact

Impact

Facilitate the analysis of the patient’s situation and enhance the recognition of severity levels.

Facilitate the analysis of the patient’s situation and enhance the recognition of severity levels.

Patient Information and Status

Patient Information and Status

Identification

A patient is identified by the number of the bad and the first letters of the name.

A patient is identified by the number of the bad and the first letters of the name.

Impact

Avoid easy recognition by visitors for privacy, as some dashboards are located in visible areas of the hospital.

Avoid easy recognition by visitors for privacy, as some dashboards are located in visible areas of the hospital.

Pending Procedures

Pending medical exams, prescriptions, or any relevant information about the patient's condition.

Pending medical exams, prescriptions, or any relevant information about the patient's condition.

Impact

Decreased the time to analyze extra information that could impact medical care.

Decreased the time to analyze extra information that could impact medical care.

Status

Status

Two steps for patient care: “Waiting for primary care” and “Waiting for the Doctor’s care.

Two steps for patient care: “Waiting for primary care” and “Waiting for the Doctor’s care.

Impact

Impact

Improved the management by showing what patients were already checked.

Improved the management by showing what patients were already checked.

Status

Two steps for patient care: “Waiting for primary care” and “Waiting for the Doctor’s care.

Impact

Improved the management by showing what patients were already checked.

Custom Protocols

Configuration of custom protocols according to the hospital's internal standard care.

Configuration of custom protocols according to the hospital's internal standard care.

Impact

Flexibility for the customers and scalability for the business.

Flexibility for the customers and scalability for the business.

Vital Signals and/or Medical Exams

Visualization of the last three vital signals results and medical exam results. In case of any change, the result number turns red to highlight the alteration.

Visualization of the last three vital signals results and medical exam results. In case of any change, the result number turns red to highlight the alteration.

Impact

Optimization of medical conduct analyses.

Vital Signals and/or Medical Exams

Visualization of the last three vital signals results and medical exam results. In case of any change, the result number turns red to highlight the alteration.

Impact

Optimization of medical conduct analyses.

Cards

The cards are organized by severity and I improved the space to include 10 patient alerts. In the small cards, I included the essential information that could impact primary care (vital signals).

The cards are organized by severity and I improved the space to include 10 patient alerts. In the small cards, I included the essential information that could impact primary care (vital signals).

Impact

Improve prioritization and empower the medical team.

Improve prioritization and empower the medical team.

Documentation

Some images of the documentation.

Impacts and Results

Since I left the company before the launch I didn’t follow the results. But here are some expected results and metrics I aimed for these projects:

Since I left the company before the launch I didn’t follow the results. But here are some expected results and metrics I aimed for these projects:

  • Reduce the time of patient care in ICUs by providing essential information to medical staff for earlier sepsis detection.

  • Reduce the progression of ICU patients to severe sepsis.

  • Increased the flow of new patient alerts due to the configuration of custom protocols.

  • Improve customer retention by empowering the users with essential information.

Get in touch!

I’m always open to chat =)

fds.andrei@gmail.com

based in Curitiba/Brazil

ALL RIGHTS RESERVED 2024 © ANDREI LEVANDOSKI

Get in touch!

I’m always open to chat =)

fds.andrei@gmail.com

based in Curitiba/Brazil

ALL RIGHTS RESERVED 2024 © ANDREI LEVANDOSKI

Get in touch!

I’m always open to chat =)

fds.andrei@gmail.com

based in Curitiba/Brazil

ALL RIGHTS RESERVED 2024 © ANDREI LEVANDOSKI

Final version of the dashboard.

Final version of the dashboard.

Final version of the dashboard.