# Study design involving 4 variables measured every month for 5 years

I want to design a study on hemodialysis patients. I know from basic science knowledge that there are four variables (suppose a, b, c, d) which effect the rates of heart attacks/ strokes and amputation in hemodialysis patients. My objective is to prove that all these have a correlation with heart attacks/ strokes and amputation in dialysis patients. After proving I want to find out which variable has the strongest link with ANY of these events. Now I have 88 dialysis patient in my hospital and have the variable data for a, b, c, d in excel form monthly for 5yrs. How can I use these variables a, b, c, d which will be 12 times in the year and 60 in 5yrs. How is time averaging done?

Let me explain more. I will only take one adverse event heart attack for now. Increase in variable A and B increases risk of heart attack and increase in variable C and D decrease risk of heart attack. How should I design a restrospective study over 5yrs to maximize the chance of finding a significance between the above mentioned hypothesis?

Variables a to d are not binary but rather continuous with limits ofcourse. Example variable C ranges from 1 to 4mg/dl in the blood. The optimum level is 4mg/dl, the lower the level the higher risk of heart attack. Now how should I analyse my data to maximize the chance of finding an association between heart attacks and low level of variable C. I will use the similar strategy for all variables. What test will I use and where can I get the software to do those tests?

I am sorry I dont know the meaning of time average. There is a study that I heard about with a similar idea. They used a variable measured every month for a few years. I am not sure how they time averaged? Perhaps it is the arithmetic mean? This is the study that uses time averaging- Dialysis Modality and Correction of Uremic Metabolic Acidosis: Relationship with All-Cause and Cause-Specific Mortality Tania Vashistha, Kamyar Kalantar-Zadeh. It appeared in the journel CJASN. The link is here- http://cjasn.asnjournals.org/content/early/2012/11/21/CJN.05780612.abstract

Is there a way to attach pdf then I can with this question. Thanks to all in advance

• Normally one "designs" a study before collecting the data: it sounds like you want to analyze study data. What information do you have about the heart attacks and strokes? What exploratory analysis of the data have you considered? What do you mean by "time averaging"? – whuber Dec 6 '12 at 8:23
• I interpret this to mean that a, b, c and d change over time - they ar enot intrinsic things for the patient (eg they are not like height, or ethnicity). One of your questions to address is whether it is a change in these variables that is linked to heart attacks etc, or the base level. – Peter Ellis Dec 6 '12 at 9:22

The values for four variables a,b,c,d (which effect the rates of heart attacks/ strokes and amputation in hemodialysis patients ) is measured per month. So, I think that the time factor should be averaged to make the value observed in per month. For example , feature a for patient A will have 12 values for one year and 60 for 5 yrs, so, all the 60 values should be summed and divided by 60 which will give a resultant value x/month. The same procedure should be followed for other features as well. At last we will have 88 rows (patients) and 4 columns (features).

• Because this proposal collapses each time series of 60 values into one mean, it seems likely it will lose a tremendous amount of potentially useful information. For instance, one would guess that changes or trends in the explanatory variables $a,b,c,$ and $d$ might be related to the endpoints, but taking the means loses all such information. – whuber Dec 6 '12 at 8:20
• Hello, thanks everyone. Let me explain. I will only take one adverse event heart attack for now. Increase in variable A and B increases risk of heart attack and increase in variable C and D decrease risk of heart attack. How should I design a restrospective study over 5yrs to maximize the chance of finding a significance? I was thinking for that I will also need to find the general event rate of heart attack in dialysis patients. How else can I calculate the power of the study or the number needed to prove an association. I only have 88 patients – sairah sharif Dec 7 '12 at 6:53
• Variables a to d are not binary but rather continuous with limits ofcourse. Example variable C ranges from 1 to 4mg/dl in the blood. The optimum level is 4mg/dl, the lower the level the higher risk of heart attack. Now how should I analyse my data to maximize the chance of finding an association between heart attacks and low level of variable C. I will use the similar strategy for all variables. What test will I use and where can I get the software to do those tests? – sairah sharif Dec 7 '12 at 6:56
• I am sorry I dont know the meaning of time average. There is a study that I heard about with a similar idea. They used a variable measured every month for a few years. I am not sure how they time averaged? Perhaps it is the arithmetic mean? I am thinking I have 12 values/yr and 60 in 5yrs, what is the best representative value? Or should I use all these values for correlating with a heart attack. How can I use all these values? What formula/ test to use? – sairah sharif Dec 7 '12 at 7:07
• Sairah, most people will not see your last three comments. Please incorporate them within your question. As far as "time averaging" goes, please provide a citatation (and preferably a link) to the study you refer to, so that we may investigate what was intended. – whuber Dec 7 '12 at 15:23