Of the multitude of survey instruments used for assessing mental health conditions, most are proprietary, expensive, redundant, onerous, diagnosis-centric, in paper form and static in the set of questions and in their scoring, based on dated data sets derived from small populations. These surveys are usually filled out in a clinic, which by itself means that the vast majority of people do not have access to them or to any insights derived from filling them out. Why can’t mental health assessment be engaging, dynamic, informative, and more inclusive, like online tools for building a personality profile?
To address this problem, we propose to build a mental health assessment framework called “My Mind Matters Quest” (M3Q) that anyone will be able to use to build a personalized mental health profile through a web browser or a smartphone app. The M3Q will be free, open source, online, adaptive, and engaging enough to entice users to come back and learn more about themselves and about other people. The Child Mind Institute (CMI) has experience creating questionnaires and administering them to, for example, 10,000 children as part of CMI’s ongoing Healthy Brain Network (HBN) study. The M3Q’s database currently includes symptoms and diagnostic categories from psychiatric statistical manuals such as the DSM-V, thousands of questions from over 80 mental health questionnaires, and information about mental health resources. We will use this database to test whether generating questions from symptoms in a principled bottom-up manner will lead to more informative and helpful answers, as per these Aims:
Most mental health questionnaires contain ambiguous questions pertaining to a limited set of symptoms. Moreover, these questions are often too extreme to allow for gradations in response from the population at large. To generate a more rigorous and inclusive set of questions, we will: (1) articulate a neutralized version of all symptoms (e.g., “failure to understand tasks or instructions” becomes “understanding of instructions”), (2) automatically apply different dimensions (frequency, duration, latency, intensity, context) to each of the symptoms as appropriate to create a set of questions (e.g., “How often do you…” vs. “How long does it take for you to…” “…understand instructions?”), and (3) reach out to CMI’s network of clinicians, psychologists, schools, etc. to crowdsource expert evaluation of our resulting set of questions.
We will evaluate the usefulness of existing questionnaires’ questions, and our derived questions, based on how well they predict diagnoses, by administering both sets of questions to thousands of diagnosed HBN participants. We will test the degree to which these predictions are governed by individual questions or by high-order interactions among many questions. Based on this, we will see how well diagnoses defined by similar patterns of answers (using either existing or automatically generated questions) match HBN clinicians’ diagnoses. Collecting questionnaire responses from thousands of diagnosed HBN participants confers the added benefit of establishing new normative standards to reflect the great diversity of mental health conditions. As more such data are added, these norms will dynamically update for use as benchmarks for diagnosis, assessment and prediction about conversion and recovery.
We will establish a framework for easy, intuitive and efficient navigation of the questionnaire. To determine an efficient, personalized path through the many possible questions, we will model the question space as an energy landscape that is searched algorithmically, as is standardly done for optimization problems in computer science. This landscape can encode the distribution of high- to low-predictor questions to help optimize efficient traversal and coverage of the mental health space. The user will be encouraged to complete their profile by invoking game mechanics that entice and reward the user to continue, such as simple feedback regarding progress, through rewards and app notifications.
The above may constitute the first framework to not only generate mental health assessment questions, but also to crowdsource their evaluation. Creating an empirically-driven, dynamically updated assessment instrument was the original intent of the DSM-V, and this project would realize that goal, to the benefit of all mental health stakeholders.