Symptom Sleuth

• HTML
• Bootstrap
• Sass
• JS
• Angular 1
• Gulp

Sep 24, 2017

Introduction

WebMD's Symptom Checker is a useful resource for patients who want to learn and get general advice about their symptoms. Surprisingly there currently is no equivalent tool for doctors and medical students that connects symptoms and test results – as they would be determined by a medical professional – to differential diagnoses and treatments. The goal of Symptom Sleuth (SS) is to address that gap: as symptoms and test results are entered, possible differential diagnoses are presented in order of match percentage. The ideal future of SS is use machine learning to train the matching algorithm to more accurately predict diagnoses given a set of symptoms.

SS was a collaboration between myself (front-end), Vincent Derelle (back-end), Dr. Bryce Burger (family physician), and Caroline Allen (medical transcriptionist).

Challenges

1With team members working in different provinces the first challenge was communication. I set up a Slack channel, integrated our GitHub (private) repo, and front-end and back-end Trello boards. In addition, Gulp and a Jenkins CI server were set up to automatically build and deploy the website. Every time code was pushed to GitHub or a task was started or completed on Trello the team was notified in Slack to check changes to the app and provide feedback.

2The input fields in the symptoms column are created dynamically depending on the "conditions" data. Conditions are fetched (currently from a local JSON file, but eventually from a database), and logic is being development to present the appropriate input fields depending on possible data input types. For instance, for some symptoms a plain text input field is sufficient, but for others the user must first be able to select the lab finding, e.g. "white blood cell count" and then be presented with the appropriate radio options, e.g. "low, normal, high". Directives have been created for these (Snippet 1) but view logic is still underway.

angular.module('symptom-sleuth').directive('ssDropdownInput', function(ssModel) {
  "ngInject";
  return {
    restrict: 'E',
    scope: true,
    templateUrl: 'views/ss-dropdown-input.tmpl.html',
    controllerAs: 'dropdownctrl',
    link: function(scope, element, attrs) {
      var subheading = attrs.ssSubheading;
      scope.listItems = [];
      
      scope.selectItem = function(item) {
        // TODO: Replaces dropdown with relevant radio inputs
        console.log(item);
      };
      
      ssModel.get().then(function(data) {
        // Create list items array for dropdown
        data.conditions.forEach(function(el) {
          if (!_.isEmpty(el[subheading])) {
            for (var key in el[subheading]) {
              scope.listItems.push(key);
            }
          }
        });
        // Sort list alphabetically
        scope.listItems.sort(function(a, b) {
          var a = a.toUpperCase();
          var b = b.toUpperCase();
          return (a < b) ? -1 : (a > b) ? 1 : 0;
        });
        // Remove duplicates from list
        scope.listItems = _.sortedUniq(scope.listItems);
      });
    }
  };
});

Next steps

SS was a challenging but rewarding endeavour. The next immediate steps would have been to

1. implement the initial (naive) algorithm for predicting diagnoses
2. add symptoms predictive symptom typing
3. set up automated framework

Addendum

Functionality ideas for SS evolve quickly, Figure 1 shows the current UI sketch.