Description of the Data

ACTG 333 was a randomized trial with a primary objective of determining whether substituting hard capsule saquinavir (SQVhc) with indivinar (IDV) or soft gelatin capsule saquinvir (SQVsgc) would show a greater decrease plasma HIV-1 RNA levels for patients with a previous prolonged (more than

Number of Error Variance Mean Absolute Predicted Error of Responses

Covariates

correlation

P=o

correlation p = 0.3

Method I Method II

Method I

Method II

0.2

0.38

0.43

0.37

0.42

10

0.4

0.50

0.59

0.52

0.59

0.6

0.68

0.74

0.62

0.66

0.2

0.37

0.47

0.37

0.53

25

0.4

0.54

0.68

0.52

0.62

0.6

0.64

0.74

0.62

0.70

0.2

0.35

0.64

0.37

0.54

40

0.4

0.53

0.76

0.49

0.65

0.6

0.64

0.80

0.61

0.81

0.2

0.37

0.69

0.38

0.64

50

0.4

0.52

0.83

0.51

0.68

0.6

0.61

0.86

0.59

0.76

0.2

0.36

1.00

0.36

0.73

100

0.4

0.53

1.12

0.52

0.83

0.6

0.66

1.16

0.63

0.95

Table 4. Comparison of the mean absolute prediction error of responses from methods I and II, assuming a normal error distribution.

Table 4. Comparison of the mean absolute prediction error of responses from methods I and II, assuming a normal error distribution.

one year) use of SQVhc. A secondary objective was to assess the predictive power of mutations in the protease gene at baseline for the in vivo anti-viral response. The mutant strains of the virus were conjectured to have developed during the prior exposure to SQVhc and could confer drug resistance. Study participants were randomized to one of the three treatment arms: 8 weeks of SQVhc, followed by IDV; 8 weeks of SQVsgc, followed by crossover to IDV if no HIV-1 RNA response; or 8 weeks of IDV, followed by crossover to SQVsgc if no HIV-1 RNA response. There were two stratification factors, one is viral load at screening (> 50, 000 or < 50,000 RNA copies/mL) and the other is the number of nucleoside reverse-transcriptase (RT) inhibitors in the anti-retroviral drug regimen (0-1 or > 2) at study entry. The original enrollment goal of the trial called for 144 participants, but the trial was stopped by the ACTG and its review board after eighty-nine subjects had been enrolled when an interim analysis demonstrated the superiority of the IDV arm.

Increased drug resistance in HIV disease has been observed with mutations leading to amino acid substitutions in the protease gene at codons 10,46,48, 82 and 84 [CC96, JHO96, VIS99] and with the accumulation of multiple mutations [CS95, CH96]. Although the HIV-1 protease gene was fully sequenced in this study, only amino acid substitutions at the 12 selected protease residues 10, 20, 24, 46, 48, 54, 71, 73, 82, 84, 88, and 90 were analyzed in the study report, because of their recognized association with resistance to SQV and/or IDV [PG00]. We explore here the use of as much as possible of baseline protease genotype, along with the treatment assignment and other baseline clinical measurements, in predicting the in vivo anti-viral response measured by the reductions in HIV—1 RNA level from baseline to week 8.

Sixty-five study subjects had measurements on HIV-1 RNA protease gene sequence. After the deletion of 5 subjects with missing CD4 measurements or detectable HIV-1 RNA level at baseline, the data set used here consisted of 60 patients who had information on protease sequence, treatment assignment, baseline clinical measurements (HIV-1 RNA viral load, the percentage of white cells that are CD4 positive (called CD4 percentile), CD4 cell counts (measured in cells/mm3), CD8 percentile, CD8 counts, prior experience with SQVhc (measured in number of weeks of therapy), and the two stratification factors. Response was defined as the reduction of HIV-1 RNA level (logio copies/mL) from baseline to week 8; a negative reduction indicated a rise in HIV-1 RNA from baseline. If the patient's RNA viral load at week 8 dropped below the quantification limit of 500 copies/mL, the corresponding observation would be right-censored. Out of the 60 observations, 12 (20%) were censored. The potential censoring value for each change in logioRNA is the difference between baseline logi0RNA and logi0 (500). Large potential censoring values correspond to subjects with high initial viral load who, because of disease burden, might respond poorly to treatment, leading to a potential dependence between censoring and response. We assume, as others have in similar situations, that including the baseline viral RNA load among the covariates mitigates this possible dependence. The analysis depends more heavily on the conditional independence of censoring and response, given the covariates, than is often the case in the analysis of censored event times from clinical trials.

Table 6 shows the distribution of the protease gene mutations among 60 subjects. One patient had no mutation, 3 had 1 mutation and the remainder had at least 2 mutations. Seventy-four codon positions had no more than 2 patients with mutations at those positions and thus were deleted as explanatory variables.

Table 5 gives the list of codon positions with at least 3 mutations. As a result, the analysis presented here used a data set of 60 subjects with 35 covariates (including 25 variables for 25 codon positions with mutations).

Was this article helpful?

0 0

Post a comment