Monday, May 20, 2013

A proposed App for smart phones to be used in primary health care in developing nations.


Here is an idea for a class of apps for smart phones. The apps would provide guidance on when and how urgently one should seek medical attention. For example, a cell phone app might help health promoters to decide when a patient should seek emergency medical attention or could wait to make an appointment for outpatient care in a few days.

Image source: Medical Informatics
The app would define a set of alternative recommendations that exhaust all possibilities. For each it would define a probability that that alternative would be recommended by an expert. The a priori probability that a patient drawn at random from the specific population (e.g. pregnant women, well babies, pre-schoolers suffering from diarrheal disease) would be taken as a baseline. The user of the app would then input data on the specific patient such as age, weight, length of current conditions, signs and symptoms. For each added piece of information that app would update the estimated probability of each alternative recommendation.

I suggest that such an app might improve recording of patient information and decision making. I have been impressed that while we assume that clinical personnel do not make mistakes, in fact they do. Use of formal probability models to improve decision making may help. Moreover, the example based on delegate medical responsibilities in a developing nation setting is consistent with the thrust of this blog, but the approach could be used by primary care physicians in the USA (on their tablet computers) or by patients with their own smart phones.

I have previously done a series of posts suggesting the procedures needed to estimate the a priori probabilities and to update them based on added information.



There might be some minor changes. For example the adjustments of probability estimates made in the above papers assume that the added pieces of information are statistically independent. In the example above, it might be that symptoms in a syndrome would not meet that requirement. The modification would probably be simple.

No comments: