Data Mining : Drug Trial Question

I have a a simple situation. I am trying to determine for how long (in months) two drugs should be given to a patient. Both the drugs are administered at frequent intervals and their dosage and the condition of the patient is recorded. How do I determine for how long should I continue to adminster the drugs,at what point should I stop? I am dealing with the following variables : Drug 1 dosage,Drug 2 dosage, Period of medication (in months). Can I use SQL Server 2005 data mining capabilites to get the answer? Please let me know.

No you cannot. An MD should evaluate the patient and decide to stop or increase/decrease dosage. Technology is yet far from of replacing an MD.|||

I have to agree with the other response that this decision should be made by a qualified MD. What data mining can do is tell you the relationships between the dosages and the medication periods. Note that since these results are dependent on data, and in this case the data is determined by the recommendations of MD's, essentially what the algorithm is doing is determining patterns in behavior of MD's prescribing drugs for an array of patients. The results could be skewed by a number of factors. For example, if all of the data came from a single MD, you would only be seeing the pattern of that doctor's prescription behavior.

What could the results be used for? I believe in this case, you could use the results for budgeting and forecasting - for example, you see the dosages and you can give an estimate to a patient as to when their treatment would end or to a payment provider as to the total cost of treatment. Another usage would be to present the patterns to MD's - e.g. as a logistic regression scorecard, tree, or some other easily consumable form - that they could use as a guideline to help them make decisions.

However, as the other responder noted, any decisions should rest in the hands of a qualified MD.

|||

Thank you, Jamie for responding to the question. I think you have very well summarized the solution, I am looking for. I am looking to find the dosages so that I can give an estimate to a patient as to when their treatment would end(an interval estimate) or to a payment provider as to the total cost of treatment (an interval estimate). Can I use logistic regression for the purpose ?

|||You could use Logistic Regression, Neural Nets, or Decision Trees. DT will give the most understandable model. I would try all three and use the accuracy chart to see which model does the best job at prediction.