topic Comparing prevalence rates across years in Statistical Procedures

Comparing prevalence rates across years

SherryBerry — Fri, 03 Jul 2015 13:58:10 GMT

Hi!

I've just started using SAS and am trying to figure out how to compare prevalence rates across years and see average percent change (95% CI).

For example, if allergy rates to Allergen A are:

2012 - 2.5%

2013- 5.6%

2014- 5.9%

What is the best way to determine the average percent change per annum?

Also, if allergy rates to Allergen B are:

2012- 1.7%

2013- 4.7%

2014- 0.7%

Is there a way to determine if there is a significant difference between rates for allergen A and B for the same year?

Thank you so much!

Sherry

Re: Comparing prevalence rates across years

Reeza — Fri, 03 Jul 2015 16:20:01 GMT

Do you have the raw data that creates those proportions? If you don't have the N's then it's difficult to

If so you can consider using the Cochran Armitage test via proc freq:

Base SAS(R) 9.2 Procedures Guide: Statistical Procedures, Third Edition

EDIT: I think I misunderstood your question, but my question still stands - do you have the raw data?

Re: Comparing prevalence rates across years

Steelers_In_DC — Fri, 03 Jul 2015 17:28:48 GMT

Here is a solution for the first part. If there is more than one record per year you would add group by year at the end. In this solution it's the average of every year.

data have;

infile cards;

input year rate;

cards;

2012 2.5

2013 5.6

2014 5.9

;

run;

data prep;

set have;

by year;

lrate = lag(rate);

diff_rate = rate-lrate;

run;

proc sql;

create table part1 as

select *,avg(diff_rate) as avg

from prep;

Re: Comparing prevalence rates across years

Steelers_In_DC — Fri, 03 Jul 2015 17:36:52 GMT

If all you want is the difference between the two averages this will get you there. I'm guessing more information is valuable you can tweak the code. Let me know if this helps:

data have;

infile cards;

input year ratea;

cards;

2012 2.5

2013 5.6

2014 5.9

;

run;

data prep;

set have;

by year;

lratea = lag(ratea);

diff_ratea = ratea-lratea;

run;

proc sql;

create table part1 as

select *,avg(diff_ratea) as avga

from prep;

data have2;

infile cards;

input year rateb;

cards;

2012 1.7

2013 4.7

2014 0.7

;

run;

data prep2;

set have2;

by year;

lrateb = lag(rateb);

diff_rateb = rateb-lrateb;

run;

proc sql;

create table prep_diff as

select *,avg(diff_rateb) as avgb

from prep2;

proc sql;

create table want as

select distinct a.avga-b.avgb as avg_diff

from part1 a inner join

prep_diff b on

a.year=b.year;

Re: Comparing prevalence rates across years

SherryBerry — Fri, 03 Jul 2015 19:36:38 GMT

Yes, I do have the raw data that created these proportions.

Re: Comparing prevalence rates across years

SherryBerry — Fri, 03 Jul 2015 19:46:01 GMT

Thank you! This got me the average difference between each year exactly like I asked. Is there any way to determine if the rates for allergen A, for example, are statistically significantly different between years? (e.g. is 2.5% different from 5.9% with a 95% CI and p-value?)

In your code above, the output shows the average change between years (avga and avgb). Is there a way to see if these averages are statistically different from each other?

Thanks again,

Sherry

Re: Comparing prevalence rates across years

Steelers_In_DC — Mon, 06 Jul 2015 13:42:15 GMT

Are you referring to the P value or is there another value that is significant for your purposes? I'm not sure I"ll be able to help much more.

Re: Comparing prevalence rates across years

SherryBerry — Mon, 06 Jul 2015 13:52:59 GMT

I'm referring to a p-value. The code that you provided gives me an annual avg percent change and I'm just looking to get a confidence interval for this change, if possible.

Thanks!

Sherry

Re: Comparing prevalence rates across years

Reeza — Mon, 06 Jul 2015 15:47:00 GMT

With 3 data points, you don't have enough data to generate a P-Value and/or CI.

Re: Comparing prevalence rates across years

SteveDenham — Mon, 06 Jul 2015 18:02:02 GMT

Since the OP has the original data, then they can do this. Any of the following could be adapted: FREQ (using Cochran-Armitage), MULTTEST (Cochran Armitage), CATMOD, GENMOD, GLIMMIX, LOGISTIC. Personally, I would probably use GENMOD or GLIMMIX with a binomial distribution. I would not fit year as a repeated measure, unless I had evidence that the same individuals were measured each year. This rules out GEE as a method (and procedure).

Steve Denham