Wednesday, November 30, 2011

A New Diagnostic Tool to Detect Depression among Cancer People

A study develops a diagnostic tool to detect depression among cancer people. Various diagnostic methods have been put forward, but all of them use heterogeneous populations without focusing on cancer patients.
Why should we develop a diagnostic tool focusing on cancer patients?
First, it’s necessary to develop a method which can identify depression in cancer patients. The reason is that depression had a negative effect on cancer patients and these depressed cancer patients have a low quality of life and higher risk of death.
Second, previous methods, which use depressive symptoms and don’t focus on cancer patients, are not suitable to deal with depression identification in cancer people. The reason is that the depressive symptoms in cancer patients may be induced by depression and other factors, such as neoplastic progression and its treatment.
What is the diagnostic tool developed in this study?
A model indicating the probability of depression of patients with cancer is constructed using the scores of some of the Ham-D items.
Which items should be selected in the model?
We want to choose a model, which can be used as a diagnostic tool with both high sensitivity (when the cancer patients are truly depressive, the score of the items in the model should be high) and high specificity (when the cancer patients are not depressive, the score of the items in the model should be low) for the diagnosis of depression among cancer patients.
What is the procedure for model selection?
First, we sample four groups of individuals, that is, patients with major depression without cancer, normal-comparison individuals, caner patients with major depression, and cancer patients without major depression.
Second, we divide the 21 items into four groups. They are “positive items”, which are endorsed by depressed patients with cancer but not by cancer patients without major depression, “common items”, which are endorsed by both cancer patients with and without major depression, “neutral items”, which are endorsed by neither cancer patients with and without major depression, “negative items”, which are endorsed by cancer patients without major depression but not by depressed cancer patients.

Third, we fit five models. Model 1 use scores of positive items, model 2 use scores of positive and common items, model 3 use scores of positive and neutral items, model 4 use scores of positive, common and neutral items, model 5, which is reference model, use scores of all 21 items.
Fourth, we plot ROC curve, which is most popular measure of the accuracy of a diagnostic test for 5 models and find that model 2 is the best.
What’s the significance for this model selection procedure?
The model selection procedure is important because this procedure can serve as a prototype to generate valid instruments for the diagnosis of major depression, even other symptoms, in a certain populations.
Which items are included in the final model?
The final model contains “positive items”, which are 6 (late insomnia), 9 (agitation), 10 (anxiety, psychic), 18 (diurnal variation) and “common items”, which are 1 (depressed mood), 14 (genital symptoms).
What’s the significance of this diagnostic tool (final model)?
Since this model has high sensitivity and specificity, it can correctly detect depression in cancer patients. Thus, depression in cancer patients can be detected and treated at an early stage.


The researchers should ensure that adequate statistical and subject-matter expertise is both applied to the study. Moreover, researchers shouldn't predetermine the outcome when they apply statistical sampling.

Source: 
[1]Guo, Y., Musselman, D.L., Manatunga, A.K., Gilles, N., Lawson, K.C., Porter, M.R., McDaniel, J.S. and Nemeroff, C.B.
The Diagnosis of Depression in Patients with Cancer: a Comparative Approach
Psychosomatics, 47: 376-384, 2006. 
Author: Lijia Wang is a first year Ph.D. student in biostatistics department. She is interested in solving problems about public health using statistical methods. She wants to improve her oral and written English.

3 comments:

  1. I'm sorry that I didn't realize the article I read was published in 2006 until I almost finished the report. (Lijia)

    ReplyDelete
  2. The idea of developing tools for specific populations is very interesting and important. I think it would have been interesting to spend time in your post discussing why that is important (you mention it some, but it would be helpful to explain it a little more). Also, it's interesting to think about the statistics behind that -- but you have to find a way to bring it down to a level that regular people understand. Even most public health people do not know what a ROC curve is, and many people will not be able to understand the picture you selected.

    Your role as a biostatistician is huge -- you don't just do the stats, but you have to find ways to explain it so that your colleagues (and sometimes even non public health audiences) can understand it.

    I am sure it is challenging to figure out what are commonly understood concepts in English when it is not your first language. : ) As you are still learning, please know that my door is always open if you want to practice and I will give you feedback on what makes sense and what is too brainy! : )

    ReplyDelete
  3. This is am important tool - thanks for sharing! I think you explain a number of concepts, which might be on my epidemiology final exam, very clearly! I wonder how this tool is being used by clinicians now (since the article was published in 2006) and whether it has improved the opportunity for detecting and treating depression earlier?

    ReplyDelete