Treatment Decisions

This table is a simple prop for thinking about treatment as opposed to observation. It could also be used to think about two competing interventions: intervention 1 and intervention 2. Along the x-axis (horizontal) in this table is the stage of the disease. The stage increases as one moves to the right as does the risk of unfavorable outcome with observation.

Treatment should be performed anywhere beyond the intersection of the outcome lines. At this point treatment improves the patient outcome (or reduces the unfavorable outcome on this chart). This can be seen only at the fifth stage when the red line crosses above the blue. The major hidden concept is that dead is dead, and it doesn't matter materially, if the patient is dead from the disease or the treatment.

This is a good model when there is good data and it is divisible into good and bad outcomes along a single dimension, but this is frequently not the case.

In the real world the loss of vision in the observation group may have a different time of onset relative to treatment. This model would require a third dimension: time. (eg;MPS subfoveal CNVM). Then one might calculate the area under the curves over time.

Frequently, the outcomes due to treatment complications are qualitatively different from the observation outcomes. For example in using Lucentis for ARMD, how does one factor an increased stroke rate against improvement in acuity measured with the number of letters. Quality of life may allow some bridging between qualitatively different categories. However, I am not always comfortable with valuation techniques.

Another situation is the decision for vitrectomy in vasculopathic diabetics with Vitreous Hemorrhage or TRD. On one side is blindness, on the other side blindness, but also of phthsis and pain requiring enucleation from surgical failure, and even loss of intellectual function and death from anesthesia.

Perhaps the even more difficult are treatments that may harbor very late complications like radiation in Retinoblastoma or chemotherapy.

The Bottom Line:
I try to simplify the analysis to this type of table as much as possible in order to clarify my treatment decisions. At the same time, I track the likely errors in my assumptions for data, complications, qualitative differences etc. Therefore, Instead of treating just beyond the intersection of the two outcome lines, I usually prefer a larger gap between the lines because the data is often biased and limited (I am particularly skeptical about all data from people or companies with money on the table, book smart geniuses, and different patient populations until it conforms with my personal observation) . In prophylaxis for retinal detachment I like a gap of 10%. On the other hand, in some diseases like Zome 1 ROP, I would treat close to the intersection because non-treatment has a very bad outcome and the progression of disease a very fast.