Example 6.1 Performing Tests with Missing Parental Data

The following data are from GAW9 (Hodge 1995) and contain 20 nuclear families that are genotyped at two markers. The data have been modified so that each mother’s genotype is missing.

     
data gaw;
   input ped id f_id m_id sex disease m11 m12 m21 m22;
   datalines;
 1    1    0    0 1 1  7  8  7  2
 1    2    0    0 2 1  .  .  .  .
 1  401    1    2 1 1  7  2  7  6
 1  402    1    2 1 1  8  2  7  6
 1  403    1    2 1 1  7  2  2  7
 1  404    1    2 2 2  8  2  7  7
 2    3    0    0 1 1  4  4  1  3
 2    4    0    0 2 1  .  .  .  .
 2  405    3    4 2 1  4  6  1  7
 2  406    3    4 2 2  4  4  3  7
 3    5    0    0 1 1  6  7  7  2
 3    6    0    0 2 1  .  .  .  .
 3  407    5    6 2 2  7  4  7  7
 4    7    0    0 1 1  1  8  7  3
 4    8    0    0 2 1  .  .  .  .
 4  408    7    8 2 2  8  4  7  3
 4  409    7    8 1 1  1  2  3  3
 4  410    7    8 2 1  8  2  7  3
 4  411    7    8 1 1  8  2  7  5
 5    9    0    0 1 1  7  1  6  2
 5   10    0    0 2 1  .  .  .  .
 5  412    9   10 2 2  7  6  6  2
 5  413    9   10 1 1  1  6  6  2
 6   11    0    0 1 1  8  4  2  3
 6   12    0    0 2 1  .  .  .  .
 6  414   11   12 1 2  8  1  2  7
 6  415   11   12 1 1  8  6  3  7
 7   13    0    0 1 1  4  6  2  2
 7   14    0    0 2 1  .  .  .  .
 7  416   13   14 1 1  4  5  2  7
 7  417   13   14 2 2  6  4  2  7
 7  418   13   14 2 1  6  5  2  6
 7  419   13   14 1 1  6  5  2  6
 8   15    0    0 1 1  6  8  2  7
 8   16    0    0 2 1  .  .  .  .
 8  420   15   16 2 1  6  2  7  7
 8  421   15   16 2 1  8  6  2  7
 8  422   15   16 2 2  6  6  7  7
 8  423   15   16 2 1  6  6  7  7
 9   17    0    0 1 2  4  7  2  7
 9   18    0    0 2 1  .  .  .  .
 9  424   17   18 2 2  4  5  7  2
 9  425   17   18 2 1  7  4  2  7
 9  426   17   18 1 1  4  5  2  2
10   19    0    0 1 1  6  4  2  7
10   20    0    0 2 1  .  .  .  .
10  427   19   20 2 2  4  4  7  2
11   21    0    0 1 1  4  7  7  7
11   22    0    0 2 1  .  .  .  .
11  428   21   22 1 1  7  6  7  2
11  429   21   22 2 2  7  4  7  2
11  430   21   22 2 1  7  6  7  3
12   23    0    0 1 1  7  6  7  5
12   24    0    0 2 1  .  .  .  .
12  431   23   24 1 2  6  4  7  7
13   25    0    0 1 1  4  1  2  8
13   26    0    0 2 1  .  .  .  .
13  432   25   26 1 1  4  8  2  6
13  433   25   26 1 2  1  8  8  6
13  434   25   26 1 1  1  4  2  6
14   27    0    0 1 1  7  6  3  2
14   28    0    0 2 1  .  .  .  .
14  435   27   28 1 1  6  2  3  3
14  436   27   28 1 1  7  4  3  7
14  437   27   28 1 1  6  2  2  7
14  438   27   28 1 1  7  4  2  7
14  439   27   28 2 2  6  2  2  7
14  440   27   28 1 1  6  4  3  7
15   29    0    0 1 1  2  4  7  4
15   30    0    0 2 1  .  .  .  .
15  441   29   30 1 1  4  2  7  7
15  442   29   30 2 2  4  8  4  7
15  443   29   30 2 1  4  2  7  5
15  444   29   30 2 1  4  2  7  5
15  445   29   30 1 1  2  8  7  5
;

Since there are missing parental data, the original TDT might not be the best test to perform on this data set. The following analysis uses the S-TDT, SDT, and RC-TDT to test markers for linkage with the disease locus.

proc family data=gaw prefix=Marker sdt stdt rctdt;
   id id f_id m_id;
   var m11 m12 m21 m22;
   trait disease / affected=2;
run;
 
proc print;
   format Probsdt probstdt probrctdt pvalue5.4;
run;

The output data set, which is created by default, is displayed in Output 6.1.1.

Output 6.1.1 Output Data Set from PROC FAMILY
Obs Locus ChiSqSTDT ChiSqSDT ChiSqRCTDT dfSTDT dfSDT dfRCTDT ProbSTDT ProbSDT ProbRCTDT
1 Marker1 5.6179 4.0083 4.7398 6 7 6 0.467 0.779 0.578
2 Marker2 12.6191 10.7500 11.9388 7 8 7 0.082 0.216 0.103

Since only one parent is missing genotype information in each nuclear family, the TDT might be applicable to some of the families. The COMBINE option can be specified, as in the following code, to use the TDT in the appropriate families, and the S-TDT or SDT for all other families. This option does not apply to the RC-TDT, so that test is omitted from this analysis.

proc family data=gaw prefix=Marker tdt sdt stdt combine;
   id id f_id m_id;
   var m11 m12 m21 m22;
   trait disease / affected=2;
run;
  
proc print;
   format Probsdt probstdt probtdt pvalue5.4;
run;

The output data set is displayed in Output 6.1.2.

Output 6.1.2 Output Data Set from PROC FAMILY Using COMBINE Option
Obs Locus ChiSqTDT ChiSqSTDT ChiSqSDT dfTDT dfSTDT dfSDT ProbTDT ProbSTDT ProbSDT
1 Marker1 4.44444 6.3692 4.2380 5 6 7 0.487 0.383 0.752
2 Marker2 2.00000 11.6489 10.7500 3 7 8 0.572 0.113 0.216

Note that the test statistics for the TDT and the S-TDT and SDT are not the same; this implies that not all families meet the requirements for the TDT. In this case, the S-TDT, SDT, and RC-TDT use more of the data than the TDT alone. However, since there is only one affected child in each nuclear family, the TDT is a valid test of association; since there is at least one occasion when there is more than one unaffected child in a nuclear family, the S-TDT and RC-TDT are not valid for testing for association of the marker with the disease locus (the SDT is always a valid test of association when the data consist of unrelated nuclear families). Both of these considerations, the amount of information that can be used and the validity for testing association, should be taken into account in deciding which test(s) to perform.

Another type of analysis can be performed using the MULT=MAX option in the PROC FAMILY statement. This option indicates that instead of doing a joint test over all the alleles at each marker, you want to perform a test to see if any of the alleles at a marker are significantly linked with the disease locus. This analysis is invoked with the following code, using only the SDT and RC-TDT:


proc family data=gaw prefix=Marker sdt rctdt combine mult=max;
   id id f_id m_id;
   var m11 m12 m21 m22;
   trait disease / affected=2;
run;
 
proc print;
   format Probsdt Probrctdt pvalue6.5;
run;

The output data set produced by this code is displayed in Output 6.1.3.

Output 6.1.3 Output Data Set from PROC FAMILY Using MULT=MAX Option
Obs Locus ChiSqSDT ChiSqRCTDT dfSDT dfRCTDT ProbSDT ProbRCTDT
1 Marker1 2.66667 2.90050 1 1 0.7173 0.6199
2 Marker2 3.57143 3.86422 1 1 0.4703 0.3946

The chi-square statistics for the tests always have one degree of freedom when the MULT=MAX option is used. Note, however, that the -values are not the corresponding right-tailed probabilities for a statistic; this is because the -values are Bonferroni-corrected in order to account for taking the maximum of several chi-square statistics.