shows that there were statistically significant differences between study groups in the proportion who were African American, and baseline BMI, CESD and FTND, due primarily to differences between populations in the two randomized studies. Since the objective of this study was to examine person-specific trajectories rather than to compare the outcomes of the different trials, we did not adjust for these differences in the calculation of the trajectory parameters. The 935 individuals in this analysis and the entire cohort of 1,155 enrollees have similar demographic and smoking history values. Continuous and 7-day point prevalence abstinence outcomes are higher for this analysis than for the full cohort, because this analysis excluded 218 individuals without symptom data, almost all of whom had dropped out and were classified as relapsers in the full cohort intent-to-treat analysis.
shows the average withdrawal symptom trajectories for the four withdrawal symptoms scales across the 7 day period. Trajectories show modest changes across time when averaged across all individuals and treatments; however, there is substantial person-to-person variability. The standard deviation of the urge-to-smoke score varies from 0.84 to 0.89 across days; the ranges for negative affect, hunger, and physical symptoms are 0.68 to 0.72, 0.84 to 0.87, and 0.36 to 0.41, respectively. shows that the standard deviation of the urge-to-smoke level is 0.70 (i.e., about 30% of individuals were more than 0.70 away from the average level of 2.8 on an urge-to-smoke scale ranging from 1 to 4).
Average withdrawal symptom trajectories
Withdrawal symptom subtype trajectory parameter descriptive statistics
shows the average trajectory parameters and their standard deviations across individuals for each withdrawal symptom scale. Person-to-person variability is substantial. For example, the between-person variability in the slope of the urge-to-smoke trajectory implies that the slope for the 5% of individuals with the largest (smallest) slopes will be approximately 1.645 × 0.0997 = 0.164 greater (less) than the average slope. Across 7 days these differences in slopes result in an additional increase or decrease in urge-to-smoke of 1 response unit (on 1 to 4 scale). While between-person variability is substantial across all withdrawal symptom subtypes and parameters, variability in the urge-to-smoke trajectory parameters tend to be as large or larger than variability in the other scales (except for variability in the hunger level).
shows the correlations of trajectory parameters within and across symptom groups. Within each withdrawal symptom cluster, the correlations of trajectory slope and curvature and curvature and level range from −.20 to −.54. For withdrawal symptom scales other than urge-to-smoke, correlations between level and volatility range from 0.28 to 0.42. Correlations of physical symptom and hunger scale trajectory features with the corresponding urge-to-smoke trajectory features are relatively modest (≤0.30), while those of negative affect trajectory features are moderately large (0.33 to 0.58).
Correlations among trajectory parameters within and across symptom groups
presents the odds ratios for logistic regressions where the dependent variables are 7-day point prevalence abstinence (PPA) and continuous abstinence at EOT and 6 months. Each regression includes the four trajectory parameters for that withdrawal symptom scale, a baseline measure of that scale obtained approximately two weeks prior to the planned quite date, demographic characteristics (African American, gender, CESD, BMI), smoking quantity and dependence (CPD and FTND), and pharmacotherapy (bupropion, placebo, NRT spray, or transdermal NRT). Odds ratios in this table are expressed as per standard deviation change in the trajectory parameters.
Odds ratios for logistic regressions of trajectory parameters on abstinence measures† in regressions including covariates‡
Prediction of abstinence
EOT— Both urge-to-smoke and negative affect trajectory features were highly predictive of PPA at EOT. The pattern of significant odds ratios was generally similar for urge-to-smoke features in the prediction of PPA and continuous abstinence whereas the consistency of the prediction pattern for negative affect features across the two outcomes at EOT was not as evident (negative affect level being the only consistently associated feature).
In contrast, to the 6 statistically significant associations for urge-to-smoke, there were only two significant associations between physical symptom features and PPA at EOT and only one for the hunger scale. None of the trajectory features for these two symptom clusters were associated with continuous abstinence at EOT. Compared to urge-to-smoke and negative affect, the trajectory parameters for these two symptom subtypes have relatively little predictive ability at EOT.
6 months—In general, the predictive power of trajectory features declined at 6 months. Urge-to-smoke and negative affect levels were both consistently associated with PPA and continuous abstinence outcomes at 6 months. Interestingly, urge-to-smoke slope and volatility were also associated with continuous abstinence at 6 months. Physical symptom and hunger trajectory parameters showed no evidence of being associated with 6-month outcomes.
presents the odds ratios for the trajectory parameters when covariates and all 16 trajectory parameters are included in stepwise regressions with backwards elimination. Only odds ratios with p values less than 0.20 are shown. Urge-to-smoke trajectory features remained significantly associated with outcomes at EOT (level, slope, and volatility) and at 6 months (level, slope, and curvature). There was only a single other statistically significant trajectory parameter (curvature for physical symptoms). Regression results were unaffected by the exclusion of baseline CESD and baseline values for negative affect and urge-to-smoke.
Odds ratios† for logistic regressions of trajectory parameters on abstinence measures‡ in stepwise regressions including all trajectory parameters and covariates§
presents the area under the ROC curve, denoted AUC, which measures the proportion of correctly classified abstinent individuals using various combinations of covariates and trajectory parameters at EOT and at 6-months. The first four rows show AUC values for various combinations of predictor variables. For example, the first row (row A) displays that when only covariates are used for classification of outcomes, the AUC varies from 0.58 (PPA at 6 months) to 0.61 (continuous abstinence at 6 months) with an average of 0.59. The AUC using covariates and urge-to-smoke scores ranged from 0.63 to 0.70 with an average of 0.67 (Row B), covariates and negative affect scores ranged from 0.60 to 0.62 with an average of 0.61 (Row C), and covariates, urge-to-smoke, and negative affect scores from 0.64 to 0.70 (Row D).
Receiver-operator characteristic AUC† values for predicting abstinence using trajectory parameters and covariates‡
The last four rows show how much the AUC values increase when various parameters are added to the predictors in rows A–D. For example, adding urge-to-smoke scores to the covariates (row E) increases the average AUC value by 0.08. Adding the negative affect scores to the covariates (row F) increases the AUC value by a modest 0.02. Negative affect has very little additive value if the regression equation already includes covariates and urge-to-smoke; in that case (row H), the AUC value increases by an average of 0.01. Adding urge-to-smoke to other predictors (rows E and G) always results in a statistically significant increase in AUC values. In contrast, adding negative affect to covariates resulted in only a single instance of a statistical significant increase (PPA at EOT; row F), and when added to urge-to-smoke never resulted in incremental statistical significance.
presents the ROC curves for predicting continuous abstinence at 6 months using (1) covariates only, (2) covariates and urge-to-smoke trajectory parameters, and (3) covariates, urge-to-smoke and negative affect trajectory parameters. Increasing distance of the ROC curve from the reference line indicates improvement in correct classification of the abstinence outcome over that expected by chance alone. We note that the curve using only covariates and the urge-to-smoke trajectory parameters is almost identical to the uppermost curve resulting from the addition of negative affect to the classification equation. A similar near overlap would occur for other abstinence outcomes.
Receiver operator characteristic curves for continuous abstinence at 6 months