arrays in stata?

[Vishal Sharma]

hi all,

I m wondering if there is a way in stata to use arrays ( like in sas) to search through multiple icd codes.

the data i have is in this format: (each line is a hospitalization)

id	date	diag1	diag2	diag3	diag4	diag5
1	20jan2018	t40.0
1	16oct2018			t40.6
2	04apr2018		t40.3

each hospitalization (each line) has the date of the event and 5 diag codes( icd codes)

i would like to identify all of the diagnosis codes that are t40.0-t40.6 such that if present, then i can create a variable pat_diagnosis=1, if absent then pat_diagnosis=0

with sas we set up an array , but with stata, i can only think of using for loops like this:

gen pat_diagnosis=0
forv i=1/5{
replace pat_diagnosis`i'=1 if diag`i'>= "t40.0" & diag`i'<="T40.6"
}

any thoughts on a better way than this?

thanks in advance
vishal

[Nick Cox]

https://www.stata.com/support/faqs/d...nt-sas-arrays/

In this particular case, your loop is confused. I think you mean

gen pat_diagnosis = 0
forv i=1/5 {
    replace pat_diagnosis = 1  if diag`i'>= "t40.0" & diag`i'<="t40.6"
}

[Weiwen Ng]

Vishal,

Quick note that Stata has two specialized commands to clean ICD-9 and ICD-10 codes, or to generate a flag if a variable contains a code in a certain range. The range feature is a nice convenience command for your purpose. You could simply do:

forv i=1/5{
icd10 generate pat`v' = diag`v', range(T400/T406)
}
egen pat_flag = rowmax(pat?)
drop pat?

I don't believe there's a direct equivalent of SAS arrays. If you needed to speed the above up, it might run faster if you reshaped the data to long (Stata command is reshape long, SAS equivalent is PROC TRANSPOSE, I believe).

[Sergiy Radyakin]

The original loop as modified by Nick is fine and looks like the most efficient solution.
If you are looking for something different, you can think along the lines of:

egen diag = concat(diag*), punct("|")
generate flag = strpos(diag,"t40.")>0

Clearly you can adjust the expression for flag to focus only on some diagnoses.

Best, Sergiy

[mark butum]

https://www.stata.com/support/faqs/d...nt-sas-arrays/

In this particular case, your loop is confused. I think you mean

gen pat_diagnosis = 0
forv i=1/5 {
replace pat_diagnosis = 1 if diag`i'>= "t40.0" & diag`i'<="t40.6"
}

What of you you have 3 or more diagnosis codes, like D45, D473 , D474. How will you create a loop to extract.
Thank you very much!

[mark butum]

I have had similar issue and Nick's code was extremely helpful, but I got stuck beyond having 3 or more codes.
Any help is extremely appreciated.

[mark butum]

This exmple gives me error:

gen pat_diag = 0
forv i=1/30 {
replace pat_diag = 1 if I10_DX`i'<="D45" & I10_DX`i'>= "D473"& I10_DX`i'<="D474"
}

[Nick Cox]

I can't see a syntax error in #7. I note that

DX`i' <= "D45"

is generously inclusive, so that anything beginning "A" "B" "C" "D1" "D2" "D3" and some things beginning "D4" would fit. What was the error message?

[mark butum]

Thanks Nick.
Sorry, I was not specific enough. I meant, when I ran above code, and tabulate results, I get
pat_diagnosis | Freq. Percent Cum.
------------+-----------------------------------
0 | 7,135,090 100.00 100.00
------------+-----------------------------------
Total | 7,135,090 100.00

[Rich Goldstein]

unlike Nick, I do see a problem with the code in #7 - you need parentheses after the ampersand (and at the end)

[Nick Cox]

Sorry; I missed something more fundamental -- as did my friend Rich Goldstein, It's impossible for a string to be BOTH earlier than or equal to D45 in alphanumeric sort order AND between D473 and D474 . Hence the syntax is legal, but no observations will satisfy the mutually contradictory criteria, as you found, You need some different mix of operators.

[mark butum]

Indeed, I am interested in adding D45, D473 and D474. I modify the code as below and seems to work

gen pat_diag = 0
forv i=1/30 {
replace pat_diag = 1 if I10_DX`i' =="D45" |I10_DX`i'=="D473" | I10_DX`i'=="D474"


pat_diag | Freq. Percent Cum.
------------+-----------------------------------
0 | 7,104,464 99.57 99.57
1 | 30,626 0.43 100.00
------------+-----------------------------------
Total | 7,135,090 100.00

Based on Nick's code, I am confident this result is correct.

My question now is, how to get separate group of patients with D45, D473, and D474.

This what I did:
gen pv=0

. forv i=1/30 {
2. replace pv = 1 if I10_DX`i' =="D45"

3. }

ta pv

pv | Freq. Percent Cum.
------------+-----------------------------------
0 | 7,130,991 99.94 99.94
1 | 4,099 0.06 100.00
------------+-----------------------------------
Total | 7,135,090 100.00



gen et=0

. forv i=1/30 {
2. replace et = 1 if I10_DX`i' =="D473"
3. }



. ta et

et | Freq. Percent Cum.
------------+-----------------------------------
0 | 7,108,341 99.63 99.63
1 | 26,749 0.37 100.00
------------+-----------------------------------
Total | 7,135,090 100.00



gen mf=0

. forv i=1/30 {
2. replace mf = 1 if I10_DX`i' =="D474"
3. }

ta mf

mf | Freq. Percent Cum.
------------+-----------------------------------
0 | 7,134,964 100.00 100.00
1 | 126 0.00 100.00
------------+-----------------------------------
Total | 7,135,090 100.00


Now I am intersted in percentages of each diagnosis:



I am not 100% confident this method I used to eventually get these percents is correct.

[ just to remind that, I looking for these 3 diagnosis: polycythemia, essential thrombocythemia, and myelofibrosis, from Diagnosis( I10_DX1 to I10_DX30). I want to capture all those diagnosis in the database. I then want to separate them into individual diagnoss with their frequencies]

I will need some reassurance or otherwise.

Thank you.

[mark butum]

i meant *frequncies

[zaed hamady]

I am wondering if someone can help me to sort out this array, where code is CD10, dates means date of this code retrieved from central registry, there are 70 codes var, and corresponding dates, in each code var there are hundreds of CD10 codes

code1	code2	code3	date1	date2	date3
c25			01/01/2015
	c25			01/02/2016
		c25			02/02/2017

Thanks
zaed

[Vishal Sharma]

thanks all!