Development and validation of a questionnaire measuring knowledge, attitudes, and current practices of primary healthcare physicians regarding road injury prevention in older adults Free

Abstract

Background

Older adults present high risk of involvement in road crashes. Preventive interventions conducted by their primary healthcare physicians (PHPC) could reduce this public health issue.

Objective

The objective of this study was to design and validate a self-administered questionnaire that measures the knowledge, attitudes, and current practices (CP) of PHCP in Spain regarding the prevention of road injuries in older adults.

Methods

One thousand eight hundred and ninety-seven PHCP completed a questionnaire piloted previously in an expert panel and two convenience samples of physicians. It comprised 78 items grouped in five sections and was mainly focused on exploring three constructs: knowledge, attitudes, and CP. Exploratory factor analysis was used to obtain evidence of internal structure validity. Reliability was assessed through Cronbach’s α coefficient. Correlation coefficients for the scores constructed for each of the extracted factors were calculated to assess convergent and discriminant validity.

Results

Factor analysis extracted four factors each for the knowledge and attitudes constructs, and three factors for the CP construct, which explained more than 55% of the variance in each construct. Except for two factors of the knowledge construct regarding existing health problems associated to the risk of involvement in road crashes, the clustering pattern of all other items across the remaining nine factors was consistent and in agreement with previous knowledge. Cronbach’s α values were greater than 0.7 for all constructs.

Conclusions

Our questionnaire appears to be valid enough to assess the attitudes, CP, and medication-related knowledge of PHCP in Spain regarding the prevention of road injuries in older adults.

Background

Increasing age has consistently been associated with a high risk of road crashes involvement and greater injury severity.^1,2 The increasing proportion of older people in developed countries’ populations³ will worsen this situation, increasing the absolute number of casualties and deaths related to road crashes among older people. Primary healthcare is the health system level where most health problems in Spain are managed.⁴ Its longitudinal nature makes this level an ideal setting to implement strategies for the prevention of crash injuries in older adults (PCIOA). However, the role of primary healthcare physicians (PHCP) in the PCIOA has yet to be clearly defined.^5,6 In order to implement a specific strategy for this purpose, it is imperative to identify the degree of knowledge, attitudes, and usual practices relevant to this issue among PHCP. Previous surveys have aimed to explore issues regarding the outcomes of primary healthcare interventions.^7–12 However, we were unable to find validated and comprehensive tools for this purpose. The aim of this study was to design and provide evidence of the validity of a self-administered questionnaire that measures the knowledge, attitudes, and current practices (CP) of PHCP in Spain regarding their role in the PCIOA.

Methods

Questionnaire design

A thorough review of the literature was conducted using metasearch engines, systematic reviews, and databases of original studies. This search guided the design of a preliminary version of the questionnaire after three previous proposals successively developed by the research team. This version was piloted in a sample of 168 family doctors at two PHCP conferences. The results of this piloting exercise were used to design and further pilot the final version of the questionnaire, in three phases:

• 1) Reformulation of the questions in the preliminary version based on the results from the pilot study, the opinion of two family doctors external to the research team and a new bibliographic review. This process resulted in an extended version of the questionnaire. Among other changes, the original five options for the Likert items were converted to four options, as we observed extremely low response frequencies for the first two of the original five categories.

• 2) Editing of the extended version. All redundant, inconsistent, imprecise, or invalid questions were removed to retain a minimum number of items. This allowed us to eliminate irrelevant variance and avoid over-representing some issues.

• 3) Testing the final version. First, it was piloted in a convenience sample of 45 PHCP from two urban and one rural primary heath care centres. Then, this version (showed in Supplementary Appendix 1), was tested in a large sample of the target population (Spanish PHCP) recruited with two strategies. First, physicians affiliated to the three main primary healthcare scientific societies in Spain were contacted by email and invited to complete the online version of questionnaire. Second, members of a consulting company trained by our team distributed the printed version of the questionnaire to physicians who attended the main PHCP conferences in Spain from October 2016 to October 2018. The number of respondents with both strategies was 1,897 physicians, 287 of which (15.13%) answered the online version. The response rate of physicians that contacted through email was 91.63% (this information could not be obtained for physicians who attended the conferences and were contacted by the consulting company). The mean time needed to complete the questionnaire was 20 min.

Questionnaire structure and content

The questionnaire included 78 items grouped in five sections. The contents of each one are summarized in Table 1. Section 1 records demographic and workplace characteristics. Section 2 (32 items) includes six questions about preventive activities in which a response of “yes” is scored as 1 and all other responses are scored as 0. A 14-item Likert-like subscale (scored from 1 to 4) explores the frequency of specific advices given to older adults. An 11-item Likert-like subscale (scored from 1 to 4) explores the level of agreement with statements regarding PCIOA-related activities. A final item asks the respondent to rank the importance of four resources for PCIOA.

Section 3 comprises a 12-item Likert-like subscale (scored from 1 to 4) addressing the level of importance of health problems potentially related with the risk of traffic accident, and three items that describe clinical scenarios and ask the respondent to choose the best clinical decision among three options. A correct answer is scored as 1 and all other answers are scored as 0.

Section 4 includes an 11-item Likert-like subscale (scored from 1 to 4) that asks how often different medications should be monitored according to their association with the risk of traffic accident, and three items that ask the respondent to choose the best clinical decision among three options that describe different scenarios involving medication use. A correct answer is scored as 1 and all other answers are scored as 0.

In section 5, the first three items, scored on a scale from 1 to 10, explore the importance physicians give to traffic accidents in older adults, along with the current and potential roles that they play in preventing road injuries. The last two items explore the respondent’s knowledge regarding traffic accidents in older adults. One point is given for the right answer and 0 for all others.

The four sections described above are designed and organized to be respondent-friendly, but not to reflect the actual internal structure of the questionnaire, which aimed to cover three constructs (also summarized in Table 1):

The CP construct aims to explore the extent to which physicians carry out PCIOA-related activities in their daily practice. This construct is covered in the first six questions in section 2, and in the first 14-item Likert-like subscale in this section (CP-subscale).

The attitudes construct aims to measure the physician’s level of motivation to participate in PCIOA-related activities. This construct is covered in the 11-item Likert-like subscale in section 2 and the first three items in section 5.

The knowledge construct aims to assess the physician’s level of knowledge about traffic accidents in older adults and their related factors, broken down in the following three dimensions: health problems (items in section 3), medications (items in section 4), and general knowledge about the importance of traffic accidents in older adults (the last two items in section 5).

Questionnaire validation

Qualitative and quantitative evidence regarding content, internal structure, convergent, and discriminant validity of the three constructs was assessed according to standard recommendations,^13–16 in the following steps:

Evidence of content validity: The final version of the questionnaire was evaluated by a panel of experts (12 members of the Department of Preventive Medicine and Public Health of the University of Granada and 17 PHCP) in order to assess the clarity, coherence, and relevance of the items, according to the three constructs.

Evidence of internal structure validity: Exploratory factor analysis applied to registries without missing values for all the items included in each construct (1,679 for CP, 1,619 for attitudes, and 1,637 for knowledge) was used to identify the underlying structure of each construct. The Kaiser–Meyer–Olkin test and Bartlett’s test of sphericity were used to verify the suitability of the data for factor analysis.¹⁷ The principal component method was used to extract the factors, and those with an eigenvalue greater than 1 were retained. Then, varimax rotation was used to make the extracted factors interpretable while keeping them separate. After items which did not fit the retained factors were removed, we built a score for each factor detected inside each construct (corresponding to each dimension), by simply adding the scores obtained for all the items included on the factor. The items included on these scores always showed uniform categories because the factor analysis did not include items with different categories in the same factor. Finally, a score for each construct was finally built by adding the scores of the factors included on it. Reliability of the retained factors and the three constructs were evaluated through Cronbach’s α coefficient.^18,19 Convergent validity was assessed using the correlation matrix between the factor scores from each construct, and discriminant validity was assessed using the correlation matrix between the factor scores for different constructs.

All analyses were done with Stata v.14 software.

Results

In our sample, 67.8% of the 1,897 physicians were women. Respondents ranged in age from 24 to 64 years, with a mean of 40.4 years (SD 12.7) and a median of 38 years (interquartile range 28–53). The mean number of years in practice was 18.6 (SD 10.3), and the median was 19 (interquartile range 10–27). Supplementary Appendix 2 shows the distribution of the remaining characteristics of the physicians.

The main conclusion of the expert panel that reviewed content validity of the questionnaire was that it was suitable for addressing all three constructs. The Kaiser–Meyer–Olkin and Bartlett’s tests provided evidence of the absence of collinearity between items. The results of the validation analyses are summarized in Tables 1–4. For the CP construct (Tables 1 and 2), the analysis identified three factors that explained 55.2% of the variance. Factor 1 (renamed CP-general advices a posteriori) included all but four items of the 14 items subscale. These four items, related with advices concerning health circumstances potentially affecting the ability to drive, were included in factor 2 (CP-health advices). Although similar loads were found for factors 1 and 2 for the item “If you need glasses or use a hearing aid, get regular check-ups,” we decided to include this item in factor 2 because it explored a type of content broadly similar to the other three items.¹⁷ Finally, factor 3 (CP-general practices) included the first six questions in section 2. Cronbach’s α for all items in the CP construct was 0.919.

The results of factor analysis for the attitudes construct are shown in Tables 1 and 3. The four factors identified explained 62.23% of the variance. Factors 1, 2, and 4 were extracted from the 11-Likert subscale in section 2. The first one (A-general) included all but five items: the three items exploring possible drawbacks to implementing PCIOA were included in factor 2 (A- drawbacks), and the two items exploring possible legal responsibilities of physicians comprised factor 4 (A-legal). Although this factor comprised only two items, we retained it in view of the coherence between them together with the large amount of variance this factor explained.¹⁷ Finally, factor 3 (A-importance) included the three quantitative items in section 5. Cronbach’s α for all items in the attitudes construct was 0.793.

Regarding the knowledge construct, preliminary analyses (detailed in Supplementary Appendix 3) lead us to discard some items and perform a final factor analysis for the remaining ones whose results are summarized in Tables 1 and 4. The 12 health problems explored in the original subscale were grouped in factor 2 and factor 4. Factor 2 (renamed K-general pathologies) included five items in this subscale (hypertension, atrial fibrillation, diabetes, obesity, and migraine). Factor 4 (K-neurologic pathologies) comprised the remaining seven items, centred on neurologic-related health problems. The items related with medication use were also grouped in two factors. Factor 1 (K-cardiovascular medications) comprised cardiovascular-related plus nonsteroidal antiinflammatory medications. Factor 3 (K-neurodepressant medications) comprised the remaining items in the medications subscale, and grouped together drugs with neurodepressant effects. The four factors explained 64.84% of the variance, and Cronbach’s α for all items retained in the knowledge construct was 0.917.

Table 5 summarizes the correlation matrix used to examine the evidence for convergent and discriminant validity of the scores constructed for each factor. Regarding convergent validity, all three scores of the CP construct were strongly correlated, especially those pertaining to the same subscale (CP-health advices and CP-general advices). In the attitudes construct, the A-general score correlated strongly with the other three scores, which were more weakly correlated with each other. Finally, all four scores of the knowledge construct correlated strongly with each other. Regarding divergent validity, although several significant correlations were found between scores pertaining to different constructs, their magnitudes were generally low. Specifically, we note the correlations between the A-general and A-importance scores with the CP-general advices score and all knowledge scores.

Discussion

In overall terms, the questionnaire we developed appears to be a useful tool for the purpose of obtaining a comprehensive view of the position of PHCP regarding PCIOA. The key outcomes are summarized and discussed below.

First, the questionnaire items clearly distinguished between the three main CP, attitudes, and knowledge constructs. We found no previous studies that defined or specifically explored the structure of these constructs, and this knowledge gap led us to perform exploratory rather than confirmatory factor analysis.^8,11,20–23 Overall, the results of our discriminant analysis supported the coverage of these three separate constructs. However, the positive correlations between some indices pertaining to different constructs pointed to a shared background, which is not surprising. It can be assumed that physicians who are highly motivated to become involved in PCIOA will tend to show higher levels of knowledge than their less motivated peers.²⁴ The need to differentiate between the three constructs is clear, given that current strategies to encourage PCIOA among PHCP (mainly procedural guidelines) have focused thus far on improving their skills and knowledge, but not on motivating them. A poorly motivated physician is unlikely to pay sufficient attention to existing recommendations and is thus unlikely to put them into practice. Moreover, current PCIOA practices by physicians may not result in a net benefit if they are not grounded on a solid knowledge base. Studies in other countries have consistently noted insufficient or nonexistent training in PCIOA.^5,24

Second, the factors associated with each construct were able to explain a large proportion of the variance. In addition, Cronbach’s α values for all three constructs were much higher than the minimum recommended value.^19,25–28

Third, the factors identified in the present analysis suggested a coherent internal structure within each construct. For the CP construct, three dimensions were clearly identifiable: general activities regarding PCIOA, general advices on safe road use, and specific advices concerning health problems frequently faced by older adults. We found no previous studies designed to consider these three practices but believe they should become the focus of further research because PCIOA requires specific interventions in addition to providing patients with general or specific advices.

The attitudes construct comprised four dimensions, two of which (A-drawbacks and A-legal) suggested that physicians’ positive attitudes towards PCIOA may be counterbalanced by concerns about both the negative effect of PCIOA on the doctor–patient relationship and the possible legal ramifications of these activities. These issues have been considered in previous research,^5,10 although we found no studies that distinguished between these perceptions and the general attitudes of PHCP regarding their role in PCIOA.

The knowledge construct was the most difficult to define and explore. In contrast to the attitudes and CP constructs, we posited three a priori dimensions (general, existing health problems, and medication use). Another difference was that a “correct” answer was pre-assigned for all items that explored knowledge. The perception by physicians that they were being tested could lead to resistance to complete the questionnaire or to provide accurate answers. Each knowledge subscales consisted of two clusters of items: those grouping health problems and medication use which, according to earlier studies,^29,30 were strongly related with lower driving skills (poorly controlled diabetes, severe obesity, sleep apnoea, peripheral vertigo, alcohol dependence, and depression among health problems, and antiepileptic agents, H1 antihistamines, codeine-based cough suppressants, benzodiazepines, muscle relaxants, and opioid analgesics among medications), and the remaining items with a weaker influence on driving ability.^31–33 The two factors identified for the medication’s subscale fit fairly well with the two groups defined a priori: K-neurodepressant medications (strong association) and K-cardiovascular medications (weak association). The only medications category with an inconclusive allocation (which led us to remove this item in the final analysis) was insulin and oral antidiabetics. However, the physicians’ answers to the health problems subscale suggested that they did not group health problems according to their actual influence (i.e. strong vs. weak) on driving ability, but according to whether they ascribed each health problem to neurological disorders based on their assumption that these problems are the most strongly related with fitness to drive. Moreover, in contrast to our expectations, the three items that explored clinical scenarios involving existing health problems and medication use were not clustered in a single factor, and this led us to remove them from the analysis of the overall knowledge construct. These results suggested that our questionnaire did not adequately capture physicians’ knowledge about the health problems most strongly related to older adults’ risk of involvement in a traffic accident.

The main limitation of our questionnaire and its validation process was the lack of previous instruments designed for the same purpose, which prevented us from consulting similar instruments to seek additional evidence of validity. We found few studies that used similar approaches to explore issues related to health problems that may interfere with safe driving, or to physicians’ attitudes regarding older patients’ fitness to drive.^9,34,35 Unfortunately, none of these previous tools has been validated.

A further limitation is information bias.^36–39 Apart from the problem regarding physicians’ perceptions of being tested, the fact that our respondents were not blinded to the purpose of the study may have led some respondents to overestimate their level of motivation and their current involvement in preventive practices.

Apart from these drawbacks, we believe that our strategy of design and validation is in line with current recommendations.^13–16 In particular, the sample we used for the validation process was large and representative enough of the target population.⁴⁰

In the development and assessment of any strategy aimed at improving the routinary performance of physicians regarding PCIOA, the use of validated tools such as the present questionnaire is essential. This should be performed in two steps: first, to define the intervention, it is necessary to know the basal conditions of the physicians; second, the tool should be used to assess its intermediate effectiveness (i.e. the improvement in knowledge, attitudes, and preventive practices of physicians).

In conclusion, except for the construct centred on knowledge of existing health problems related with the risk of older drivers’ involvement in a traffic accident, the validity of our questionnaire appears to be solid enough to assess the attitudes, CP, and medication-related knowledge of PHCP in Spain regarding PCIOA. Further research is needed to identify potential variations in validity for specific subgroups within our target population, and in other potentially relevant target populations, e.g. other healthcare workers, specialized and tertiary levels of healthcare, or professionals in other countries.

Acknowledgements

We thank all participants for their valuable contributions to this study, and K. Shashok for improving the use of English in this manuscript.

Funding

This work was supported by the SEMERGEN-UGR Chair of Teaching and Research in Family Medicine (Cátedra de Docencia e Investigación en Medicina de Familia SEMERGEN-UGR), University of Granada, Spain.

Ethical approval

All relevant ethical guidelines have been followed.

Conflict of interest

The authors declare that they have no conflicts of interest.

Data availability

The data underlying this article will be shared on reasonable request to the corresponding author.

References