An  applications for computers in medicine

I think about what I do as a physician and a specialist in hematology and oncology,  and how it could be made automatic. If it we're automatic, it would be done more uniformly and rigorously and the calculations would be more reliable.

I believe that machines can do many  things better than I can. Their memories are  perfect,.  Their calculations are exact.

I also believe that, at least at this point, I can add something to the machine. I can be more skeptical. I can understand the motivations of the people who create the input. I can understand the implications of suffering, the feelings of relief and loss.

I look forward to a collaboration with the machine. I do my best to foster it. I have brought the internet to clinical conferences for the last 30 years. Do not forget the sounds of the telephone modem connection in the 1980's.

I would like to see an expanded role for computers in medical decisions. .The  Electronic Medical Record was supposed to help with this integration. It has done a very poor and limited job. Sometimes, it has stood in the way of using the capabilities of computers to help analyze and solve problems. It is very difficult to export  data to a spreadsheet for analysis. This will undoubtedly get better. But the structure of incompatibility among electronic medical records is a violation of one of the main motives of the government that encouraged,  with sizable economic incentives,  their adoption.

In my hematology practice, I am often confronted with the question of the cause and treatment of low blood counts. I have found an analysis of the time course of the fall of such blood counts is useful. Often, a point in time can be identified when the fall began. A pattern can be seen suggesting either a slow, gradual fall or a rapid fall or both phenomena superimposed upon one another. Although these consultations are a welcome source of income for me, I recognize that the algorithm that I use could be made  automatic.  Such an automatic algorithm would help me do this work. Currently, an automatic solution would need to be checked by a human health care provider. Over time, the level of training of that provider might not require subspecialty certification. The primary care provider could do an adequate job of spotting unusual circumstances

I would like to see the machine identify the time course of variations in laboratory tests and then, automatically, correlate these with variations in treatment. This is not a hard algorithm. Thinking about it was somewhat entertaining. I would do it by going back to the definition of the derivative. One would then look at the slopes of the segments that make up the overall change and identify the more dramatic slopes as times of dramatic change to be correlated with alterations that might be identified in the history, especially medication changes. ( After doing the calculations, I realized that this was probably  the beginning of Newton’s method of difference,  the origin of calculus that I had recently read about in James Gleick’s biography of Isaac Newton.)  

This approach could also be used to monitor whether interventions are effective. Improvements could be correlated with interventions directed at those parameters.

Such an algorithm does not replace the human It simply helps her. I think we need it.

Model curve

"derivative" curve identifying days of greater and lesser change