Explaining clinical behaviors using multiple theoretical models

Background In the field of implementation research, there is an increased interest in use of theory when designing implementation research studies involving behavior change. In 2003, we initiated a series of five studies to establish a scientific rationale for interventions to translate research findings into clinical practice by exploring the performance of a number of different, commonly used, overlapping behavioral theories and models. We reflect on the strengths and weaknesses of the methods, the performance of the theories, and consider where these methods sit alongside the range of methods for studying healthcare professional behavior change. Methods These were five studies of the theory-based cognitions and clinical behaviors (taking dental radiographs, performing dental restorations, placing fissure sealants, managing upper respiratory tract infections without prescribing antibiotics, managing low back pain without ordering lumbar spine x-rays) of random samples of primary care dentists and physicians. Measures were derived for the explanatory theoretical constructs in the Theory of Planned Behavior (TPB), Social Cognitive Theory (SCT), and Illness Representations specified by the Common Sense Self Regulation Model (CSSRM). We constructed self-report measures of two constructs from Learning Theory (LT), a measure of Implementation Intentions (II), and the Precaution Adoption Process. We collected data on theory-based cognitions (explanatory measures) and two interim outcome measures (stated behavioral intention and simulated behavior) by postal questionnaire survey during the 12-month period to which objective measures of behavior (collected from routine administrative sources) were related. Planned analyses explored the predictive value of theories in explaining variance in intention, behavioral simulation and behavior. Results Response rates across the five surveys ranged from 21% to 48%; we achieved the target sample size for three of the five surveys. For the predictor variables, the mean construct scores were above the mid-point on the scale with median values across the five behaviors generally being above four out of seven and the range being from 1.53 to 6.01. Across all of the theories, the highest proportion of the variance explained was always for intention and the lowest was for behavior. The Knowledge-Attitudes-Behavior Model performed poorly across all behaviors and dependent variables; CSSRM also performed poorly. For TPB, SCT, II, and LT across the five behaviors, we predicted median R2 of 25% to 42.6% for intention, 6.2% to 16% for behavioral simulation, and 2.4% to 6.3% for behavior. Conclusions We operationalized multiple theories measuring across five behaviors. Continuing challenges that emerge from our work are: better specification of behaviors, better operationalization of theories; how best to appropriately extend the range of theories; further assessment of the value of theories in different settings and groups; exploring the implications of these methods for the management of chronic diseases; and moving to experimental designs to allow an understanding of behavior change.

of patient, type of visit e.g. 6 monthly recall), Patient (age, anxiety, cooperative behaviour, attendance, attitude to teeth, attitude to dentist). From this five clinical scenarios were constructed describing patients presenting in primary care with dental caries (see Additional File 1). All should receive a restoration.
Respondents were asked to decide whether or not they would do a restoration and decisions in favour were summed to create a total score out of a possible maximum of five.

Behavioural intention
Three questions assessed intention to restore: 'I aim to use restorations to manage caries in children under 17 years of age; I have in mind to use restorations when I see children under 17 years of age; I intend to restore teeth as a primary part of managing caries. Responses were summed (range 3 -21) and scaled so that higher scores reflected greater intention to restore. Procedure An independent statistician, using a list of random sampling numbers, selected 450 dentists from the Scottish Dental Board practice list. These dentists were sent an invitation pack (letter of invitation, questionnaire consisting of psychological and demographic measures and a consent form to allow access to their fee claims data from MIDAS, as well as a reply-paid envelope). Three postal reminders were sent to non-responders at 2 weeks, 4 weeks and 6 weeks from the first mailing (April to June 2004). Routinely collected data on fee claims for treatment, used to generate the primary outcome measure, were gathered for a 12 month period (6 months before and after the first questionnaire posting) to control for seasonal variations.

Sample size and statistical analysis
The target sample size was based on a recommendation by Green 6 to have a minimum of 162 subjects when undertaking multiple regression analysis with 14 predictor variables and an expected response rate of approximately 40% from previous surveys of this population. The overall analytic approach was to first check the internal consistency of the measures. Where necessary questions were removed to achieve a Cronbach's alpha of 0.6 or greater. Where this was not possible the highest alpha was achieved. For two question constructs a correlation coefficient of 0.25 was used as a cut off. Next, for each of the three outcome measures, Pearson Correlation Coefficients between the individual constructs and the outcome measures were calculated and then multiple regression analyses were used to examine the predictive value of each theoretical model. For the five "perceived cause of illness" questions in the Common Sense Self-regulation Model responses were dichotomized into scores of five to seven (indicating agreement that the cause in question was responsible for caries) versus anything else (indicating disagreement). These dichotomous variables were then entered as independent variables into the regression. Finally, for predictors that were statistically significant, irrespective of whether or not they came from the same theory, we examined the relationship between predictive and outcome variables. All constructs which predicted the outcome were entered into a stepwise regression analysis to investigate the combined predictive value of significant constructs across all theories. The relationship between predictive and outcome variables were examined using ANOVA for the Stage Model. The relationship between Implementation Intention and intention was not explored as Implementation Intention is a post-intentional construct.

Ethics approval
The study was approved by the UK South East Multi-Centre Research Ethics Committee. 2 (IQR 1 to 3). Median number of sessions worked per week: 8 (IQR 9 to 10). Table 1 contains descriptive information on the behaviour data models. The descriptives are the means of the totals of the score of the items in the construct. As this is linear regression and we present beta, it doesn't make any difference (the information in a mean of means and a mean of totals is identical). A mean of means puts all the constructs on the same scale (apart from those that have a multiplicative component) making them easier to interpret.

Description of dataset
In the tables, have used the * p< 0.05, ** p <0.01, *** p <0.001 throughout. r = Pearson correlation between the construct and the outcome variable (either behaviour, simulation or intention). Beta = the standardised regression coefficient, positive means a positive relationship, negative a negative relationship, the size of the coefficient measures the size of the relationship between the construct and outcome, adjusted for the other constructs in the model, as a shift in proportion of a standard deviation in the outcome for a shift of a standard deviation in the construct. The adjusted R 2 column gives the proportion of the variance explained by the model, the other columns after that are information in the regression (F statistic, and numerator and denominator degrees of freedom). The descriptive text in the table describes the constructs.
Theory of planned behaviour, none of the constructs were predictive of behaviour, no variance in behaviour was explained by this theory. In fact this was the case for all theories except the knowledge items The higher the knowledge score, the lower the behaviour, (there was a negative correlation between knowledge and behaviour), knowledge explained 5% of the variation in behaviour (this is from the adjusted R squared from linear regression). Table 2 contains descriptive information on the behaviour simulation (BS) and intention data models. Intention was correlated with BS, as was the attitude indirect construct. The regression model with intention and PBC direct predicting BS predicted 5.3% of the variance in BS, adding in PBC power construct did not add to this model significantly (and PBC power was only weakly correlated with BS). Social cognitive theory explained 13.1% of the variation in BS, the outcome/expectancy and self efficacy constructs were the significant variables. Action planning explained 3.7% of the variation in BS. Evidence of habitual behaviour was the only predictive construct from operant learning theory; the theory explained 5.9% of the variance in BS. Self regulation model, and the knowledge construct did not predict BS. Stage theory explained 7.3% of the variance in BS.
Theory of planned behaviour predictor variables explained 27.9% of the variance in intention, the two attitude constructs were significant, the rest weakly correlated with intention. Social cognitive theory explained 21.4% of the variance in intention, as with BS, it was outcome/expectancy and self efficacy constructs were the significant variables. Implementation intention (action planning) explained 24.5% of the variation in intention. Operant learning theory explained 51% of the variation in intention, all the constructs were bivariately correlated with intention, but in the regression model only evidence of habitual behaviour and experienced consequences predicted intention (because anticipated consequences is correlated with both the other constructs in this theory.). Self regulation model explained 18.8% of the variance in intention. Stage theory explained 13.0% of the variance in intention. Knowledge items did not predict intention. Table 3 shows the cross theory analysis. For behaviour the only significant predictor was knowledge so the results are the same as the individual theory section Table 1). Outcome/Expectancy and Self Efficacy together explained 14.2% of the variation in BS. Evidence of Habitual Behaviour, Attitude direct, Action Planning, Outcome expectancy, Anticipated consequences, Experienced consequences explained 61.6% of the variation in intention. -0.250** -0.250** 0.05 1, 114 7.6** *p= or <0.05; ** p= or <0.01; ***p= or <0.001. (a) Only intention and perceived behavioural control measures are entered into the regression equation as only these constructs are the proximal predictors of behaviour in this model. Alpha = Cronbach's Alpha; r = Pearson product moment correlation coefficient; Beta = standardised regression coefficients; -= single question measure. a) PAP stages were distributed as follows ; unmotivated 96(74%) , motivated more 1 (1%) , motivated less 4 (3%), action more 8 (6%), action less 19(15%) , not responded 2 (1%); ANOVA analysis showed that Behavioural stage did not predict the number of restorations performed: F(2,111) = 0.66, p = 0.521. Note: the stages have very small cells for an ANOVA; stage theory has 3 stages, unmotivated, motivated and action. These Stage theory results for behaviour are from ANOVA on percentages presented: behaviour F(4, 109) = 0.352, p = 0.842.