Most Popular

    Sorry. No data so far.


Winner of Amgen Patients | Choices | Empowerment Competition Emerging Star of HealthCare Engagement Award
Mayo Clinic Award - LeftA winner of the Mayo Clinic iSpot Competition for Ideas that will Transform HealthcareMayo Clinic Award - R

The Patient Decision Cycle

How can patients’ decisions be optimized? I didn’t know, but I was determined to find out. At a recent Stanford MediaX workshop on Augmented Decision Environments that I was lucky enough to attend, Neil Jacobstein described a continuously improving decision cycle used by the military and based on over 50 years of decision theory.

The steps of the cycle?

Observe -> Assess the situation -> Determine the objectives -> Generate alternate plans -> Project probable outcomes -> Select the best plan -> Communicate and implement the plan -> Validate and improve the model -> back to Observe… and around again, improving the decision process with each loop around the cycle.

In a breakout group, I fleshed out how to apply this cycle to patients, with help from Neil, Matt Butner, and another wonderful contributor who prefers to remain anonymous. We implemented it (above) as a patient user experience flow design.

We are publishing this as an open framework for anyone out there who is helping patients navigate their health. Here is our thought process through the cycle:

A patient has a health concern.

1. Health Data (Observe) – The first step is to gather data from as many sources as possible to get the best observation of the patient. Self-reported data, Personal Health Records, Electronic Medical Records, genome sequence, biosensors, environmental and pscyhological tests are all welcome, and passive data recording is preferred.

2. Probable Diagnosis (Assess the situation) – Next the data is passed through various analyses, from clustering and simple correlations to modeling and more complex Artificial Intelligence methods. The more data collected, the better the models can be. Pooling with known datasets may be one way to reduce error and bias.

3. Personal Goals (Determine the objectives) – As the system learns more about the lifestyle and values of the patient, it feeds this information into a Netflix-like recommendation engine. For example, are they seeking to maximize weight loss or minimize pain?

4. Treatment Engine (Generate alternate plans) – Based on the data, diagnosis, and goals, a series of custom, personalized recommendations is made to the patient including drugs, specific lifestyle/diet changes, and alternative therapies to consider.

5. Probable Outcomes (Project probable outcomes) – As with probable diagnosis, statistical models are applied here to show the patient what she can expect by following each of the recommended treatment plans. Visualizations projecting smoking cessation or decline in quality of life over different timeframes help the patient decide how aggressively she wants to go forward.

6. Personal Plan (Select the best plan) – This is where the patient makes the decision to move ahead on one specific plan. She signs up for the online program with visual analytics to help her see her progress, and the real-world program where smart devices, medications, meals, and resources are delivered to her.

7. Support and action (Communicate and implement the plan) – To help keep her on track, a patient chooses to share her progress with real-time support in social communities, both online and in person. A robotic or virtual personal medical assistant provides reminders and encouragement, and keeps the patient up-to-date on her deliveries of needed supplies.

8. Assessment (Validate and improve the model) – At periodic intervals, the patient checks in and the system matches her current health data to her goals. If necessary, course corrections are suggested and the cycle is iterated again.

I’d love to hear feedback on this model, in the comments below or at alexandra@curetogether.com.


You can follow any responses to this entry through the RSS 2.0 feed. You can leave a response, or trackback from your own site.

AddThis Social Bookmark Button

5 Responses to “The Patient Decision Cycle”

  1. Interesting model, thanks for sharing with the community!

    Can you explain more about the purpose of the model and how it is used?

    Your examples of each step seem to describe a vision of a Health2.0 world with technology involved in every step of the decision-making process. I can see how this is a useful map in understanding how the various technologies, products, and services in development today will fit together to serve the patient in the future. (It would be fun, insightful, and educational to overlay companies/techs to see what role each will play; it may also indicate gaps representing opportunities for new offerings).

    I can also see how this model can be applied to explain how patients make decisions today. In this case, most of the answers to the questions at each step are answered for the patient by her physician (e.g. the physician explains the treatment options and probable outcomes to the patient based on experience, training, and recent clinical studies). I think mapping out this process would also be insightful and useful for explaining the weak links in our healthcare system–from the user’s perspective! Which leads to my second question.

    2. How does this model account for other people involved in the decision-making process?

    A model that included the physician in this process could help provide guidance in how the patient-physician relationship can be optimized, making the most of the physician’s expertise while improving health for the patient.

    [It would be fun to start a collaborative presentation with the model and see how it could be built upon. I'm thinking of something like prezi.com.]

  2. Thanks Jeremy, great comments! Yes, it would be fun and useful to map existing technologies, companies, and other people in the decision-making process onto this map. I should clarify also that this is not necessarily how patients are making decisions TODAY, but how patients can optimally make decisions. So moving towards a model like this can help improve health outcomes for everyone.

    Awesome idea to use Prezi, I’ve started it here – http://su.pr/3RIZI6

    If anyone wants to collaborate on building the model, let me know and I’ll send you a Prezi invitation.

  3. I would like to see an example of this model in use. For example, a patient with problem X went through such-and-such assessment and set the following goals, etc, etc, and this is what happened. I think it’s really interesting but would like to see it’s practical application.

  4. Agreed that applying this model to real-world situations is the ultimate test of its value. Also, Neil Jacobstein replied to my request for feedback with the following comment (thanks Neil!):

    ————-

    Nice job! I have two specific suggestions:

    1. The Step 3 description says: “Personal Goals (Determine the objectives) – As the system learns more about the lifestyle and values of the patient, it feeds this information into a Netflix-like recommendation engine.” However, a Netflix-like engine, independent of the movie domain, is not a good model for recommending health practices. Specifically, it is a sophisticated clustering algorithm, that makes recommendations based on your movie (or health) preferences and the preferences of people it matches with you. However, unlike movies, which can be about anything real or unreal, interacting with our ancient biological systems is not just a matter of preferences. Even if thousands of people like you are doing x in specific situation y, that doesn’t make it safe or effective. There are many examples of crowds being unwise about health practices. Consider that the key objective in Step 3 is human health, and the subobjectives may be returning specific physiological parameters such as weight or TSH levels to values currently considered more optimal.

    2. The Step 4 description says: “Treatment Engine (Generate alternate plans) – Based on the data, diagnosis, and goals, a series of custom, personalized recommendations is made to the patient including drugs, specific lifestyle/diet changes, and alternative therapies to consider.” This step is possible at some imperfect level, but still fraught with uncertainty and risk. The medical data are changing as new experiments are done, the patient’s personal health stats are changing dynamically (consider Stafford Beer’s lagging indicators diagram), patients vary genetically and with respect to acquired allergies, there are still many unknowns, there are even more uncertainties. We want to be able to toss patient data into a giant web of medical knowledge and always get cogent and accurate advice out, but the current state of the art is quite far behind this worthy research goal. I have been working indirectly towards this kind of goal for a very long time, and think it is important to acknowledge current limitations. To paraphrase an old saying: “clarity of vision is not proximity to goal.” Yes, patients or “participants” may check the web page box with all the appropriate disclaimers, but they will expect (read demand) far more perfection than the current state of the art can deliver. You might consider adding some qualifiers to the description in Step 4 and others.

  5. In an ideal world, I think doctors would have some visibility to each of these steps. In a world of constrained resources, and one with the inertia of the contemporary American system, I would locate doctors in the cycle adjacent to: “Diagnosis” and “Action”. I see two merits of this approach.

    First, Medicare DRG codes, our best template for bundling payment to align doctor’s incentives, are tied to “diagnoses”. Accordingly, so are some existing billing codes. Billing codes also exist for the “action” steps.

    Second, if we want to move clinical care towards “focused factories” (e.g., Herzlinger), bifurcating diagnosis from treatment allows appropriate specialization. A trend in medical tourism, for example, is to have the diagnosis in the rich world and the treatment in the 3rd world. It also avoids problems with gain sharing between the diagnosing physician and the treating physician.

    P.S. Alexandra if you want to send me access to the Prezi we can see how these look bolted on.

Leave a comment or question