Information Technology Reference
In-Depth Information
set nis.charlsoncopd;
pr=pr13;
run;
data nis.p14;
set nis.charlsoncopd;
pr=pr14;
run;
data nis.p15;
set nis.charlsoncopd;
pr=pr15;
run;
data nis.procedureforcopd;
set nis.p1 nis.p2 nis.p3 nis.p4 nis.p5 nis.p6 nis.p7 nis.p8 nis.p9
nis.p10 nis.p11 nis.p12 nis.p13 nis.p14 nis.p15;
run;
Most (but not all) of the procedures in Table 7 are related to COPD. However, the most frequently
occurring procedure is still only performed for just under 8% of the patients.
All of the procedures are rare occurrences since they occur in 8% of the COPD population or less.
Table 8 shows the most frequent procedures for a primary diagnosis of COPD. We find the procedure
codes by modifying the previous code slightly:
data work.charlsoncopd2;
set nis.charlson;
if (rxmatch('486',dx1)> 0 ) then code= 1 ;
else code= 0 ;
data nis.charlsoncopd2;
set work.charlsoncopd2;
where code= 1 ;
Usually, once the codes are identified, indicator functions are defined as 1=the diagnosis is present
for the patient and 0=the diagnosis is not present. Then the indicator functions are used in a regression
equation. Whether the number of codes is 20 or 50, not all potential diagnosis codes can be used; we
must limit them in some way.
However, the use of indicator functions requires two additional assumptions when used in regres-
sion. First, it must be assumed that all providers code these conditions uniformly the same way. That is,
for X i =(X i1 ,X i2 ,…,X in ) where X i represents patient i and X i1 ,…,X in represent the values of the indicator
functions; X i and X j must be independent and identically distributed. Without the uniformity of data
entry, these values are not identically distributed. Another problem occurs if the indicator functions are
not independent; that is, if co-morbid conditions are related since the multicollinearity will inflate the
variance of the model. If collinearity exists, very likely, the variance, standard error, and parameter es-
timates are all inflated. In other words, the high variance is not a result of good independent predictors,
but a mis-specified model that carries mutually dependent and thus redundant predictors. However, it
is well known that many patient factors are co-morbid and related. Consider Tables 9 and 10 showing
Search WWH ::




Custom Search