topic Confidence Interval as >999.999 in PROC LOGISTIC in SAS Procedures

Confidence Interval as >999.999 in PROC LOGISTIC

Sathish_jammy — Mon, 08 Apr 2019 08:41:59 GMT

Dear Experts,

I have trouble in finding the confidence interval value for the attached (CSV) data. Please have a look at the attachment.

The data contains 2 variables Grade (1 / 0) - dependent variable, WeightHeightRatio value(Independent variable).

While evaluating an odd-ratio for the data I get a result as kind of infinity. ---->999.999 ????????

Odds Ratio Estimates and Profile-LikelihoodConfidence Intervals

Effect	Unit	Estimate	95% Confidence Limits
WHtR	1	440.666	72.929	>999.999

proc logistic data = reg2 plots=roc;
model Grade1(event='1') = WHtR;
roc 'WHR' WHtR;
run;

/*I tried a alternate way to ommit the infinity value*/
proc logistic data = reg2 plots=roc;
model Grade1(event='1') = WHtR / FIRTH clodds = pl;
roc 'WHR' WHtR;
run;
/*But get the same result*/

I attach my dataset as well here. Please let me know where I went wrong.

Thanks in advance!

Re: Confidence Interval as >999.999 in PROC LOGISTIC

Rick_SAS — Mon, 08 Apr 2019 16:07:44 GMT

You have not necessarily done anything wrong. The estiamte for odss ration is EXP(b1), where b1=6.1056 is the estimate for the WHtR variable. The lower/upper limits are

exp(b1 +/- 1.96*StdErr), which is

72.784571

2761.9604

For a discussion and example, see the end of the article "Formats for p-values and odds ratios in SAS."

Re: Confidence Interval as >999.999 in PROC LOGISTIC

Sathish_jammy — Thu, 11 Apr 2019 08:56:48 GMT

Dear @Rick_SAS

Thank you so much for your suggestion.

But it looks like weird to get a very big odds ratio in this logistic regression.

Do you think, Is it fine to considerable to present in the report?

Even though I followed a few procedures to get exact results.

1. Removing Outliers

2. Amount of Missing Values and handle the missing values

but still, the odds ratio is huge.

Odds Ratio Estimates

Effect Point Estimate 95% WaldConfidence LimitsWHtR

448.348

72.779

>999.999

Analysis of Maximum Likelihood Estimates

Parameter DF Estimate StandardError WaldChi-Square Pr > ChiSqInterceptWHtR

1	-1.5159	0.5301	8.1777	0.0042
1	6.1056	0.9276	43.3206	<.0001

As per your statement :

The estimate for odds ratio is EXP(b1), where b1=6.1056 is the estimate for the WHtR variable. The lower/upper limits are

exp(b1 +/- 1.96*StdErr), which is

72.784571

2761.9604

Can I present an Odds Ratio as 448.348 (95%CI 72.78 - 2761.96) ...

Re: Confidence Interval as >999.999 in PROC LOGISTIC

Rick_SAS — Thu, 11 Apr 2019 09:03:05 GMT

> Do you think, Is it fine to considerable to present in the report?

You should discuss this issue with your colleagues or supervisor.

> Can I present an Odds Ratio as 448.348 (95%CI 72.78 - 2761.96)

Yes, you could do that. If you use four significant digits, it would be (72.78, 2762).