# Compare disease code between two years

Data:

Name    Dx_2013 Name    Dx_2014
A   25060   A   25070
A   35720   A   44381
A   25000   A   16100
A   41390   A   25060
A   49600   A   35720
A   25200   A   49600
A   16100   A   72500
A   44290   A   35940
A   40390   A   24490
A   41400   A   27240
A   58520   A   41400
A   24490   A   44000
A   27240   A   25000
A   40190   A   40190
A   44000   A
A   16190   A

B   16290   B   19850
B   19850   B   19910
B   19910   B   41200
B   25000   B   49600
B   40110   B   16290
B   41200   B   26390
B   49600   B   58530
B   16230   B
B       B   49120
B       B   25000
B       B   19889
B       B   19820

C   31100   C   40110
C   25000   C   25002
C   40110   C
C   53081   C
C   49600   C
C   49390   C

D   25063   D   24490
D   25073   D   25063
D   27801   D   25073
D   35720   D   35720
D   41200   D   41200
D   42830   D   41390
D   44020   D   44381
D   44381   D   44020
D   79902   D   49600
D   25001   D   27240
D   24490   D   31100
D   24961   D   40190
D   27240   D   42830
D   31100   D   53081
D   40190   D
D   44390   D
D   53081   D
D   41401   D

E   13500   E   13500
E   42731   E   25000
E   27801   E   27801
E   24490   E   40190
E   25000   E   42731
E   27200   E   29600
E   27240   E   24490
E   40190   E   25002
E   27220   E   27200
E   41390   E   27240
E   41401   E   29680


Data is given for 5 patients, between year 2013 & 2014 for the diseases patients visited the hospital for.

The numbers are numerical codes for diseases,called Dx code (5 digits)

The diseases are chronic, once the diseases were identified, these will remain in the patient, regardless of whether he treated the condition in the future or not.

The challenge here is, we see for Patient A, he had treated 16 Dx in 2013 but only 14 Dx in 2014, likewise for other patients, he may not have been treated for conditions he had in 2013, or he may have developed new conditions altogether in 2014

I need to understand the diseases between years by using some statistical measures

What would be the appropriate measure to use, to understand the Dx's between years for a given patient?

Is there any correlation measures or measure of agreement that we can use?

The code from 2013 could change to something else in 2014, Example: code 25000 is for diabetes if a person had diabetes in 2013, he may have the same code in 2014, or an advanced diabetes codes such as 25010, 25020 up to 25090. likewise for other diseases

In cases where codes change between years what should be the measure to use.

E 13500 E 13500 E 27801 E 27801 E 24490 E 24490 E 25000 E 25000 E 27200 E 27200 E 27240 E 27240 E 40190 E 40190 E 42731 E 42731 E 27220
E 41390
E 41401
E E 29600 E E 25002 E E 29680

I have re-arranged data for patient E, such that i see patient E had treated his 2013 conditions in 2014. He did not treat 3 of his 2013 conditions (27220, 41390, 41401) and developed 3 new conditions in 2014 (29600, 25002, 29680). I want to map these Dx codes between years, 1) exactly 2) if condition advanced in 2014 then i should match Dx 25000 with Dx 25010.

• It is not clear what exactly is your research question. Is it that you want to determine difference in prevalence of diseases between 2 years or is it what disease change in persons between 2 years? You have to define it clearly before planning your analysis.
– rnso
Apr 20, 2015 at 12:41
• I need to compare 2014 Dx codes with my 2013 Dx codes for each patient. The challenge in the data set is that, you will not find all Dx in 2014 which were in 2013, and from 2013, the disease may have become complicated requiring a change in code, say from 25000 to 25010, in such a situation i should be able to answer questions 1. Whether all of my 2013 Dx codes present in 2014? 2. If a patient increased his complexity, warranting a change in code, how to capture that? 3. Is there any statistical measure that i can use to correlate my Dx's between two years? Apr 21, 2015 at 7:07

You can ask what proportion of disease codes of patients remained same in 2014? (Remaining proportion would have changed). And statistical significance of the same.

So you need to determine, for each patient, the number of codes in 2013 and then how many (& proportion) of these changed at 2014. Means of these numbers will give estimate for whole data. Tests of proportions can be applied for statistical significance.

For the data given above, following seem to be the initial values:

NUMBER OF DISEASES FOR EACH SUBJECT:

  Name Dx_2013 Dx_2014 diff percent_diff
1    A      16      14   -2    -12.50000
2    B       8      11    3     37.50000
3    C       6       2   -4    -66.66667
4    D      18      14   -4    -22.22222
5    E      11      11    0      0.00000


SUMMARY OF COLUMNS:

             Dx_2013   Dx_2014      diff percent_diff
length      5.000000  5.000000  5.000000      5.00000
min         6.000000  2.000000 -4.000000    -66.66667
max        18.000000 14.000000  3.000000     37.50000
median     11.000000 11.000000 -2.000000    -12.50000
mean       11.800000 10.400000 -1.400000    -12.77778
sd          5.118594  4.929503  2.966479     37.69753
se          2.289105  2.204541  1.326650     16.85885


Paired t-test: between Dx_2013 and Dx_2014

data:  Dx_2013 and Dx_2014
t = 1.0553, df = 4, p-value = 0.3508  <<<<<<<<<<<<<<<<<<<<<<<
alternative hypothesis: true difference in means is not equal to 0
95 percent confidence interval:
-2.283371  5.083371
sample estimates:
mean of the differences
1.4

• Thanks rnso, I am not interested in total number of Dx for person A in 2013, compared with 2014, instead i am trying to see whether Dx in 2013 is getting treated in 2014 or not. For example i want to see whether dx code 25000 in 2013 got treated in 2014 either as 25000 or with an advanced code of 25010. Thanks for your help. Apr 23, 2015 at 10:30
• You should write in your question what is your desired output on the sample data that you have posted. Write at least for one subject so that we know what exactly you want to extract from the data.
– rnso
Apr 23, 2015 at 11:31
• E 13500 E 13500 E 27801 E 27801 E 24490 E 24490 E 25000 E 25000 E 27200 E 27200 E 27240 E 27240 E 40190 E 40190 E 42731 E 42731 E 27220 E 41390 E 41401 E E 29600 E E 25002 E E 29680 I have re-arranged data for patient E, in such a that i see patient E had treated his 2013 conditions in 2014. He did not treat 3 of his 2013 conditions (27220, 41390, 41401) and developed 3 new conditions in 2014 (29600, 25002, 29680). I want to map these Dx codes between years, 1) exactly 2) if condition advanced in 2014 then i should match Dx 25000 with Dx 25010. please let me know if this helps. Apr 23, 2015 at 12:20
• Add these to your question so that it is clearly seen by all. So these are for subject E. I believe you want 2 columns.
– rnso
Apr 23, 2015 at 12:26
• yes correct, i will add this to my question. Apr 23, 2015 at 12:28