Decision making in health care involves consideration of a complex set of diagnostic, therapeutic and prognostic uncertainties. Medical therapies have side effects, surgical interventions may lead to complications, and diagnostic tests can produce misleading results. Furthermore, patient values and service costs must be considered. Decisions in clinical and health policy require careful weighing of risks and benefits and are commonly a trade-off of competing objectives: maximizing quality of life vs maximizing life expectancy vs minimizing the resources required. This text takes a proactive, systematic and rational approach to medical decision making. It covers decision trees, Bayesian revision, receiver operating characteristic curves, and cost-effectiveness analysis, as well as advanced topics such as Markov models, microsimulation, probabilistic sensitivity analysis and value of information analysis. It provides an essential resource for trainees and researchers involved in medical decision modelling, evidence-based medicine, clinical epidemiology, comparative effectiveness, public health, health economics, and health technology assessment.

Thanks, Mohammad. Yes, it is indeed complex when it comes to determining the best course of action scientifically, and I was wondering if simple things like incidence and prevalence count in the day to day practice of a clinician. For instance, if the incidence of dengue is high in the area a doctor is practicing, would he/she be more inclined to order a dengue rapid test for atypical symptoms such as fever that lasted for a few days for his patients even in the absence of other characteristic symptoms? My question is, do doctors or nurses find it useful to know the frequency of occurrence of diseases in their population when diagnosing and treating patients?

Thanks Saigopal for the nice tool! It is very useful to calculate the PPV and related values.

Just curious, would you know of any clinicians who does include these considerations in his/her clinical practice?

Also, in that vein, are doctors in your country generally aware of the prevalence of diseases in the country and actively use that information to assess whether some tests are useful or not? 

I wonder if it is possible to consider all these things carefully when clinic consultations are generally about 5-10 minutes. 

Clinically relevant prevalence and sensitivity/specificity data is fairly rare.

Hyperlocal factors may have a dramatic impact on prevalence. 

Take COPD as an example - 

Imagine you are a physician working in a city with 3 hospitals

1.  A large inner city hospital associated with a medical school

2.  A medium sized suburban hospital

3.  A Veteran's Affairs (VA) hospital

Will all the hospitals have the same prevalence of COPD?

Probably not. 

Where do you find published data to answer this question?

I would suggest you try this for a large city of your choice.

Most physicians develop an informal estimate of the prevalence of a condition in the population they work with and tailor diagnostic strategies with this information.

I do not think that clinicians, at least general practitioners, generally are very much aware of incidence and prevalence of diseases and disorders. They often think that "common" and "increasingly common" is reflected in how often they see patients with various conditions in their own clinic. Sorry! Not the answer you were hoping for.

@Dr Robinson: Thanks! Yes, I have observed that doctors usually develop a rough idea about the kinds of symptoms and associated diseases in the population they see the most. Perhaps prevalence and incidence is really more relevant to policy making resource allocation.

@Prof Leboeuf-Yde: Thanks for your frank insight! I did observe that but was not sure if it was the case.

You can read the following recent paper:

Lee WC, Wu YC. Characterizing decision-analysis performances of risk prediction models using ADAPT curves. Medicine 2016;95(2):e2477.

The paper describes an "ADAPT curve" which neatly characterizes the decision-analysis performances of a diagnostic test or a risk prediction model.

Clinical decision making such as diagnoses and differential diagnoses derive strength from information about incidence and prevalence. Example; in trying to unravel the diagnosis behind a febrile illness, apart from history taking, physical exams and lab investigations, the clinician also considers existing information on incidence and prevalence of the numerous causes of febrile illness in that particular area of practice before arriving at a definitive diagnosis and also listing the possible differentials. Listing malaria as a possible differential diagnosis in Germany (for instance) is completely out of place because it is not supported by any global data on prevalence or incidence of malaria...etc etc

Hi Sajogpal,

Thank you for that informative elaboration on the critical role prevalence estimates, in any setting, play in determining the utility of any diagnostic test.

Regards,

Physicians clearly do use the prior probability that a patient has a given disease. Even if they don't formally use "naive Bayes" or such, and even if they don't quantify the prior probability consciously, they obviously do use it by for example ordering tests for the most likely conditions first.

The question is then how useful prevalence and incidence are as surrogates for the prior probability. One problem is that the literature values of prevalence and incidence don't apply directly to a specific patient, and they will almost always be averages across broader groups, and for the specific patient the physician will have additional relevant information such as symptoms, already ruled-out alternative disorders, and even just the mere fact that the patient went to see the GP.

Then there's the issue with prevalence that it measures the probability of diagnosed disease, while what the doctor wants to know is the probability of not-yet-diagnosed disease. And with incidence it is even worse: it's the probability of getting diagnosed within one year, but who knows how long a disease is diagnosable before it actually gets diagnoses? If the answer is "three years" then you would have to multiply the incidence rate by 3.

So it's a good question. My guess would be that they are potentially better than no information at all, but at the same time there's a big danger of misinterpretting their implication for clinical practice.

Obviously they are easier to use in settings where we don't have additional patient-specific information such as when designing populations screening programmes.

There are some clinical decision models that incorporate incidence rates as predictors, for example

Fine AM, Nigrovic LE, Reis BY, et al. Linking surveillance to action: incorporation of real-time regional data into a medical decision rule. J Am Med Inform Assoc 2007;14:206–11

the number from hospital or center of statistic in any country it should be studied to know the health situation on country