Methodology used for further studies 4.1 Study background
4.13 Statistical analysis
4.13.1Software: SPSS
Quantitative data analysis was performed using Statistical Package for Social Scientists (SPSS version 12.0.1). Data obtained from the questionnaires A, B, C and practice record extraction were in the form of dichotomous (nominal and ordinal) variables and continuous variables. Spirometry test and calibration data were in the form of continuous variables.
4.13.2 Presentation of data
The distribution of continuous variables was explored visually using stem and leaf plots. The Kolmogorov-Smirnov and Shapiro-Wilks tests of normality were applied. For continuous variables with a normal distribution, results are presented as mean and standard deviation (SD). Where variables are non-normally distributed, results are presented as median and interquartile range (IQR).
4.13.3 Statistical tests
Decisions on the methods of data analysis are discussed separately for each research question.
4.13.3.1 Comparison of dichotomous variables
The Chi squared test was used to compare dichotomous variables between groups and Fisher’s exact test was used if any of the expected values in the comparison were less than five. Alpha level was set at 0.05 and p values of < 0.05 were taken as indicating significant differences.
4.13.3.1.1 For TN and UC practices variables compared were:
Gender, smoking status, current respiratory diagnosis, reported use of respiratory medications.
Number and quality grades of spirometry tests performed in target group participants Classification of spirometry by study algorithm
Proportion of participants with OLF or RLF having data extraction Recorded use of respiratory medications both before and after spirometry New medication recorded
Medication change recorded
Recording of repeat spirometry, referral for respiratory specialist consultation, chest radiology and pulmonary rehabilitation.
Recording of respiratory symptoms, exacerbations and physical activity and smoking status
4.13.3.1.2 For NLF, OLF and RLF groups variables compared were: Gender, smoking status, current respiratory diagnosis and medication use Proportion of participants with FER <0.7
4.13.3.1.3 For TN practices variables compared were: Number of participants agreeing to or declining spirometry
Gender and smoking status of participants agreeing to or declining spirometry
4.13.3.1.4 For those with AO (FER<0.7) variables compared were:
Gender, smoking status, current respiratory diagnosis, and current use of respiratory medications.
Severity of airflow limitation
4.13.3.1.5 For smokers in NLF and OLF groups variables compared were: Gender, smoking history, current respiratory diagnosis, current use of respiratory medications, attaining education post grade 12, presence of dyspnoea on exertion, proportion living with a partner,
Making quit attempt > 24 hours within past 12 months MRC functional dyspnoea grades
Proportion consulting GP following spirometry Self-efficacy and social support for quitting Smoking status at follow up
Quit attempts made during 3 months
Stage shifts- forwards, backwards, no change
4.13.3.1.6 For smokers who quit and continuing smokers variables compared were: Gender
MRC functional dyspnoea grades TTM stage
GP record of smoking status and smoking cessation advice GP record of new respiratory diagnosis and medication
Banded scores for self-rated general health, lung damage and benefits of stopping smoking
4.13.3.2 Comparison of paired categorical outcomes
McNemar’s non-parametric test for two related dichotomous variables was used to compare dichotomous nominal variables measured before and after spirometry.
4.13.3.2.1Variables compared were:
Use of respiratory medications pre- and post-spirometry
4.13.3.3 Continuous variables with normal distribution in two-group comparison Two independent groups were compared using Student’s T-test. Levene’s test for equality of variances was applied and if homogeneity of variables was found, the unequal variance estimates of significance were used. The paired T-test was used to compare repeated measures of a variable. A result was considered significant where the p value <0.05 and 95% confidence intervals excluded zero.
4.13.3.3 1 For TN and UC practices variables compared were: Age, smoking history
4.13.3.3 2 For participants with/without AO (FER<0.7) variables compared were: Age, smoking history
4.13.3.3 3 For smokers in NLF and OLF groups variables compared were: Heaviness of Smoking Index
4.13.3.3 4 For smokers who quit and continuing smokers variables compared were: FEV1 % predicted
4.13.3.3 5 For spirometer calibration checks using a random compared to the dedicated spirette variables compared were:
Deviations from 3-litres for expiratory checks and inspiratory checks (paired test).
4.13.3.3 6 For spirometer calibration checks using a random or the dedicated spirette variables compared were:
Absolute expiratory and inspiratory volumes Inspiratory and expiratory calibration test volumes
4.13.3.4 Continuous variables with normal distribution in three-group comparison Means of variables for more than two groups were compared by a one-way analysis of variance (ANOVA) for variables that met the assumptions of population normality and homogeneity of variance.
4.13.3.4 1 For target group participants in NLF, OLF and RLF groups variables compared were:
Age
4.13.3.5 Continuous variables with non-normal distribution
The non-parametric Mann-Whitney test was used for comparisons between two independent groups when variables were not normally distributed.
4.13.3.5 1 For target group participants in TN and UC practices variables compared were:
FEV1 % predicted, pack years smoking,
Frequency of consultations pre-spirometry and post- spirometry
4.13.3.5 2 For participants with/without AO variables compared were: Self-rated general health and lung damage
4.13.3.5 3 For smokers in NLF and OLF groups variables compared were: Age, age of starting smoking, number of cigarettes smoked per day, pack years smoking, % predicted FEV1
Self-rated general health, lung damage and benefits of stopping smoking
4.13.3.5 4 For smokers who quit and continuing smokers variables compared were: Age,
Pack years smoking, Heaviness of Smoking Index, Number of cigarettes/day Self-rated general health, lung damage and benefits of stopping smoking
4.13.3.6 Paired continuous outcomes with non-normal distribution
The Wilcoxon signed rank test was used for comparison of repeated measures of variables that were not normally distributed and takes into account the magnitude of the differences between the two paired variables.
4.13.3.6.1 For record review in TN and UC practices variables compared were: Number of consultations pre-and post-spirometry
4.13.3.7 Continuous variables with non-normal distribution in three-group comparison
The Kruskal-Wallis non-parametric test for comparison of more than two
independent groups was used to compare differences for variables that violated the requirements of normality of distribution and homogeneity of variance required for ANOVA.
4.13.3.7.1 For target group participants in NLF, OLF and RLF groups variables compared were:
Pack years smoking
Self-rated general health and lung damage FEV1 % predicted
4.13.3.7.2 For smokers in different stages of change at baseline variables compared were:
Self-rated general health, lung damage and benefits of quitting
4.13.3.8 Agreement on spirometry classification
Cohen's kappa was used for agreement between spirometry evaluation using the study algorithm by the trained nurse and a physiologist.
4.13.3.9 Agreement between two measures
The Bland-Altman plot is method of comparing the agreement between two measures of a clinical outcome (272). A plot of the difference between the
measurements against their mean is displayed graphically. The limits of agreement are set by the requirement for measured calibration volumes not to differ by more than 3.5% from 3-litres (113).
4.13.3.9.1 For calibration checks with dedicated or random spirette variables compared were:
Measured and target inspiratory and expiratory calibration volumes by calibration flow and time in use
4.13.3.10 Correlation- non parametric procedure
Correlation between categorical variables with a skewed distribution was investigated using Spearman’s rank order test.
4.13.3.10.1 For smokers in the OLF and NLF spirometry feedback: Self-rating for general health and FEV1% predicted
Self-rating for general health and lung damage Self-rating for general health and quit benefit Self-rating for general health and smoking exposure Self-rating for lung damage and quit benefit
Self-rating for quit benefit and smoking exposure
4.13.3.10.2 For smokers who quit and continuing smokers:
Self-rating for general health, lung damage, benefits of smoking cessation and FEV1% predicted.
4.13.3.11 Logistic regression
Regression models are a category of statistical model that describe mathematically the dependence of one variable on one or more other variables (273). Logistic regression analysis is one type of multivariable modelling procedure that is used to assess the relationship between two or more continuous or categorical, explanatory (independent) variables and a single binary outcome variable such as mortality or morbidity (smoking or quit smoking) (274,275). In linear regression, the relationship between the predictor and response variables is described by the equation:
ŷ = α + βx
where ŷ represents the predicted value on the dependent variable, x represents an individual value on a predictor variable, α corresponds to the intercept of the regression line and β the slope of the regression line (coefficient).
The mathematical model for logistic regression is similar, but equations can be developed to predict probabilities for either of the dichotomous outcomes. The ratio of probabilities for the dichotomous outcomes is the odds ratio, obtained from the antilogarithm of the coefficient in the regression equation (276). The odds ratio represents a change in the estimated odds of the outcome compared to the reference group or when the continuous variable increases by one unit. Since odds ratios are an accepted measure of association in medical research, logistic regression analysis was used in this study to explore the effect of the independent variable of interest,
spirometry feedback on sustained smoking cessation and making at least one quit attempt lasting more than 24-hours three months post-spirometry (275). Logistic regression analysis allows the flexibility to use continuous explanatory variables, that may not be normally distributed (e.g. self-rated beliefs about health) and
dichotomous variables such as gender. Where necessary, continuous variables were transformed using bands into ordered categorical variables. Tests of linear trend were performed with ordered categorical variables using the category level as a covariate. Interactions, in which the impact of one variable depends on the level of another variable, were examined in the data and are reported where significant interactions were found (275). The percentage for the dependent variable that was correctly identified by a logistic regression model was assessed by reporting the value of
4.13.3.12 Multinomial logistic regression
Multinomial logistic regression is an extension of binary logistic regression when the categorical dependent outcome has more than two levels (278). This analysis
computes separate equations for the comparison of outcome categories against a designated reference category. The outcome of interest according to the study
hypothesis was stage shift in the Transtheoretical Model. There were three categories for this outcome: forward shift, no change or backward shift. No stage change was designated as reference category. The effects of the independent variable of interest, spirometry feedback, and other variables thought or known to influence smoking cessation, were investigated in univariate and multivariate analyses. Interactions between spirometry feedback and perceived health beliefs were examined.