topic Re: Trend test for Incidence rates over the years in SAS Procedures

Trend test for Incidence rates over the years

smunigala — Thu, 12 Nov 2015 18:10:47 GMT

Hi,

I need advice about how to calculate a significance test (trend test) for comparing the incidence rates over the years. I need to comment on whether there is a significant increase or decreas (with p-value for a trend) for incidence rates over the years.

Please see the attached data. My denominator for incidence is 100,000 for all.

I would really appreciate if anyone can help me with the SAS code.

Thanks,

Satish

Re: Trend test for Incidence rates over the years

Steelers_In_DC — Thu, 12 Nov 2015 19:05:40 GMT

You should give more information regarding the calculations. While I'm sure there are several that know exactly what you want I do no.

Re: Trend test for Incidence rates over the years

PGStats — Thu, 12 Nov 2015 21:56:31 GMT

What have you tried, so far?

Re: Trend test for Incidence rates over the years

smunigala — Thu, 12 Nov 2015 23:23:06 GMT

Thanks for the reply. The incidence rates are the trends for acute pancreatitis deaths nationally over the years. I wanted to know if there is an increase or decrease in trend over the years. The incidence rates are adjusted for 2010 population (national US population).

Re: Trend test for Incidence rates over the years

smunigala — Thu, 12 Nov 2015 23:25:24 GMT

I have not done the trend test for comparing rates over the years to check for significance. However, I did Cochran- Armitage Trend test previously for categorical variable. So I am not sure how to do this. I looked up some online blogs about linear regression analysis but not entirely sure. I hope this helps.

Thanks!

Re: Trend test for Incidence rates over the years

PGStats — Fri, 13 Nov 2015 04:04:19 GMT

A popular test for detecting trends in timeseries is the Kendall Tau-b correlation, available in proc corr and proc freq :


proc corr data=trendtest kendall;
var year; with amortality_pan;
run;

Re: Trend test for Incidence rates over the years

PGStats — Fri, 13 Nov 2015 04:28:23 GMT

Did something happen around 2001?

If so you should test before and after separately:

libname xl Excel "&sasforum.\Datasets\Trendtest.xlsx" access=readonly;

proc sql;
create table trendtest as
select *, year>=2001 as period from xl.'Sheet1$'n;
quit;

proc sgplot data=trendtest;
scatter x=year y=amortality_pan;
refline 2001 / axis=x;
run;

proc corr data=trendtest kendall;
by period;
var year; with amortality_pan;
run;

Re: Trend test for Incidence rates over the years

smunigala — Tue, 24 Nov 2015 20:46:24 GMT

Hi PGStats,

Thanks for the response. I am not sure if there was anything significant in 2001 but recently over the years, the trends decreased due to improved diagnostic tests.

Satish

Re: Trend test for Incidence rates over the years

Totoro — Wed, 08 Feb 2017 22:26:59 GMT

Hi, I have a similar question like this. After doing this step, how could I interpret the 0.4463? What is the trend for the data? Since the prevalence or the incidence are not normally distributed, could I just use Simple Regression Model and get the coefficient for the trend? Or is there any other suggestion for this situation? I have a sample data like below. Thanks a lot.

Kendall Tau b Correlation Coefficients, N = 41 Prob > \|tau\| under H0: Tau=0
	Year
Prevalence	0.44634 <.0001

Year	Prevalence
1974	16.943
1975	19.6
1976	18.934
1977	18.48
1978	19.924
1979	12.369
1980	17.302
1981	15.954
1982	16.75
1983	17.184
1984	17.722
1985	19.008
1986	18.452
1987	16.744
1988	17.029
1989	17.643
1990	17.179

Re: Trend test for Incidence rates over the years

PGStats — Thu, 09 Feb 2017 05:45:29 GMT

You will get a lot more attention if you submit your question as a new topic.

You can get a robust estimate of the slope called Sen's slope (or the Kendall slope estimator) as:

proc sql;
create table slope as
select 
    median( (b.prevalence-a.prevalence) / (b.year-a.year) ) as senSlope
from
    myData as a inner join 
    myData as b on a.year < b.year;
select * from slope;
quit;

Re: Trend test for Incidence rates over the years

smunigala — Thu, 09 Feb 2017 15:32:16 GMT

HI PGStats,

Can you please respond to my other post "Trend Test for Incidence"

Thanks you!

Re: Trend test for Incidence rates over the years

Totoro — Mon, 13 Feb 2017 20:16:00 GMT

Thanks a lot. It works and report a number 0.189. Could I interpret 0.189 as each year the prevalence is increasing 0.189 (the annual year of change). Is that correct? I really appreciate for your help.

Re: Trend test for Incidence rates over the years

PGStats — Mon, 13 Feb 2017 21:52:42 GMT

I( guess so. But I haven't seen your code, or your full dataset. Make sure you understand what you are doing.