Cross-sectional and longitudinal analyses of nutrition guidance by primary care physicians

Transcription

1 European Journal of Clinical Nutrition (1999) 53, Suppl 2, S35±S43 ß 1999 Stockton Press. All rights reserved 0954±3007/99 $ Cross-sectional and longitudinal analyses of nutrition guidance by primary care physicians GJ Hiddink 1 *, JGAJ Hautvast 2, CMJ van Woerkum 3, MA van't Hof 4 and CJ Fieren 5 1 Dairy Foundation on Nutrition and Health, Zonnebaan 12D, 3606 CA Maarssen, the Netherlands; 2 Department of Human Nutrition and Epidemiology, Wageningen Agricultural University, Wageningen, the Netherlands; 3 Department of Communication and Innovation Studies, Wageningen Agricultural University, Wageningen, the Netherlands; 4 Department of Medical Statistics, University of Nijmegen, Nijmegen, the Netherlands; and 5 Primary Care Physician, Wageningen, the Netherlands Objective: To investigate in primary care physicians (PCPs) the determinants of a nutrition guidance practice (`noticing patients' overweight and guidance of treatment'), as well as their mechanism of action, in a crosssectional and a longitudinal approach. Design: Mixed longitudinal design. Five years follow up study of a previous cross-sectional study in October Subjects: A representative sample of 675 Dutch PCPs, in practice for 5 up to 20 y. Interventions: A shortened version of the Wageningen PCPs Nutritional Practices Questionnaire was mailed to the subjects in August Main outcome measure: To obtain with the LISREL-program a model of the mechanism of action of determinants of the dependent variable `noticing patients' overweight and guidance of treatment' with an adequate t of the empirical data, both in the cross-sectional and in the longitudinal approach. Results: The same set of predisposing factors and intermediary factors explains the dependent variable both in two different representative cross-sectional study populations of PCPs, and in a cohort cross-sectional study at two points in time. Two dynamic LISREL-models were developed (the `determinant-longitudinal approach' and the `early behaviour longitudinal approach') which explain the dependent variable. The latter model has, as added value, a gain in explained variance. In 5 y time, the dependent variable decreased signi cantly (P < 0.001). Conclusions: This study recon rms that PCPs' nutritional guidance practices are determined partly directly by predisposing factors, and indirectly via driving forces and barriers. However this study also reveals that an important nutrition guidance practice of PCPs, `noticing patients' overweight and guidance of treatment', shows a signi cant decrease over the last 5 y. At the same time, two of the four predisposing factors and two of the three driving factors also decreased signi cantly. As research ndings indicate that the role of diet in health and disease becomes of greater in uence PCPs need to be activated to apply their responsibility in this eld within a multi-faceted approach. Sponsorship: The study was supported by the Dutch Dairy Foundation on Nutrition and Health, Maarssen, The Netherlands. Descriptors: primary care physicians; nutrition guidance; determinants; mechanism of action; LISREL; longitudinal research; The Netherlands Introduction Physicians are perceived as the best source of health information, the most credible source and after the media the source most often used (Worsley, 1989; Hiddink, 1992, 1996; Hiddink et al, 1997c). For nutrition information primary-care physicians (PCPs) can potentially play a key role (Hiddink, 1992). In 14 ± 28% of consultations diet *Correspondence: GJ Hiddink, Dairy Foundation on Nutrition and Health, Zonnebaan 12D, 3606 CA Maarssen, The Netherlands. Contributors: GJ Hiddink, JGAJ Hautvast, CMJ van Woerkum, MA van't Hof and CJ Fieren designed the study. GJ Hiddink was the principal investigator. He reduced the Wageningen PCPs Nutritional Practices Questionnaire for use in this study and organized together with the Market research bureau GfK, Dongen, The Netherlands the data collection. MA van't Hof and GJ Hiddink are responsible for statistical data-analysis. GJ Hiddink wrote the rst draft and the nal version of the manuscript. JGAJ Hautvast, CMJ van Woerkum, MA van't Hof and CJ Fieren contributed to the preparation of the manuscript and to the interpretation of the results. GJ Hiddink is the guarantor for the integrity of the article as a whole. comes up for discussions (Worsley, 1989; van Dusseldorp et al, 1988; Murphy & Vogel, 1984), the initiative being evenly divided between PCP and patient (van Dusseldorp et al, 1988). Because of their high referral scores, high perceived expertise and reach to nearly all segments of the population, PCPs are in a unique position compared with dieticians and the Netherlands Food and Nutrition Education Bureau, the two best competitors out of eleven sources of nutrition information in the Netherlands (Hiddink et al, 1997c). The involvement of PCPs in nutrition guidance practices however appears to be very low (Orleans et al, 1985; Levine et al, 1993). Levine et al observed that clinical practices of PCPs related to nutrition are well below a minimum level as de ned by Young et al (1983), and found that favourable attitudes of PCPs towards using nutrition guidance in their practice were not consistent with PCPs' clinical reports (Levine et al, 1993). During the rst International Workshop `Nutritional attitudes and practices of primary care physicians' (December 1995, Heelsum,

2 S36 The Netherlands; proceedings published as a Supplement of the American Journal Clinical Nutrition, June 1997), Glanz reviewed the extent of nutrition counselling in the United States and in Europe (Glanz, 1997). Most studies focused primarily on counselling for high blood cholesterol and obesity and revealed that 50 ± 75% of PCPs report conducting some nutrition counselling in their practice (Wiesemann, 1997; Glanz & Gilboy, 1992; Glanz et al, 1995; Levine et al, 1993). According to a recent study, some aspects of nutrition (for example excess energy consumption and consumption of a balanced diet) are considered to be less important than they were a decade ago (Wechsler et al, 1996). In general, there appears to be a trend toward more frequent nutrition counselling since the early 1980s (Glanz, 1997). PCPs also rely on dieticians and nurses to provide nutrition counselling and report they themselves usually spend no more than 5 min discussing dietary change at a single visit (Glanz et al, 1995). Glanz & Gilboy reviewed nine studies of cholesterol management and counselling in which data were collected with observations, chart audits and surveys of patients. They found rates of nutrition counselling ranging from 17% to 70% (Glanz & Gilboy, 1992). The determinants of nutrition guidance practices of PCPs are poorly understood. So far, studies searching for determinants of nutrition guidance practices of PCPs have been limited to identifying perceived barriers of PCPs as discussed before (Hiddink, 1996; Hiddink et al, 1997b) or by addressing speci c areas, such as cardiovascular risk reduction (Mann & Putnam, 1989; Ammerman et al, 1993; Glanz & Gilboy, 1992). In a previous cross-sectional study (Hiddink et al, 1995), determinants of nutrition guidance practices of PCPs were identi ed (Hiddink et al, 1997a), as well as their mechanism of action (Hiddink et al, 1997b). Nutrition guidance practices of PCPs are directly and signi cantly based on a few predisposing factors: driving forces and perceived barriers may act as signi cant intermediary variables. The question now is whether these ndings show a sustainability in other cross-sectional studies and in time (a cohort cross-sectional study). Answering these questions is essential in order to understand how PCPs practise nutrition guidance of their patients and with what kind of interventions this nutritional guidance can be made more effective (Green & Kreuter, 1991). In the present paper we address these questions by means of a cross-sectional and a longitudinal analysis. As nutrition guidance practice we studied the variable `Noticing patients' overweight and guidance of treatment' (Hiddink et al, 1997a). Methods In October 1992, a random sample of 1000 primary care physicians (PCPs) was drawn from the population of Dutch PCPs who had been practising for 5 up to 15 y. They received a specially developed mail questionnaire (the Wageningen PCPs Nutritional Practices Questionnaire), based on the methodology of Dillman (Dillman, 1978) as fully described previously (Hiddink et al, 1995). The 633 respondents (net response rate 64%) were well representative of the population of PCPs who had been in practice for 5 up to 15 y according to gender, year of starting practice, and gender by type of practice distribution (Hiddink et al, 1995). In the current study, a shortened version of the Wageningen PCPs Nutritional Practices Questionnaire (obtainable for the rst author) was mailed in August 1997 to a nationwide random sample of 675 Dutch PCPs, who had been in practice for 5 up to 20 y. This random sample was achieved in the following way: Six hundred of the 633 respondents from the rst study were eligible. They were supplemented with 222 PCPs who had been in practice for 5 up to 10 y Ð on the basis of gender by type of practice by health center (yes=no) distribution Ð to become representative for the total Dutch population of PCPs in practice for 5 up to 20 y. (Database Netherlands Institute of Primary Health Care (NIVEL): data per 1 July 1997; addresses and information on sex, type of practice, year of starting practice and degree of urbanization of the practice). In order to assure independent self-completed questionnaires, an equal chance was given to every PCP, and only one member of each (combined) practice appeared in the sample. This sample of 822 PCPs was divided in two representative samples of (675 PCPs and of 147 PCPs) on the basis of gender by type of practice by health center (yes=no) distribution. The sample of 675 PCPs was used in the present study, which was carried out along a standardized protocol. 371 PCPs responded to the questionnaire, 297 males and 74 females (Table 1). Forty- ve percent of PCPs had a solo-practice, 37% a dual-practice, 11% a group-practice and 7% worked in a health center. Twenty-one percent of GPs practices were in rural areas and 79% were in urban areas. Information about non-participants was based on the same NIVEL-data. 283 GPs were part of the longitudinal study (in practice for 10 up to 20 y) (Table 1). Dependent variable `noticing patients' overweight and guidance of treatment' The variable `noticing patients' overweight and guidance of treatment' was chosen as a typical example of nutrition guidance. This variable was operationalized in six items (Hiddink et al, 1997b). One item addressed the percentage of patients of whom the PCP noticed their weight. Five items were about guidance of treatment: three of these concerning the discussion of overweight problems and two items concerning the extent of the advice. In 1997 we identi ed the mechanism of action of determinants of nutrition guidance practices of PCPs by means of linear structural relationship analysis (LISREL) using a postulated model (Hiddink et al, 1997b). The postulated model on the mechanism of action was con rmed. The model Table 1 Design of studies of 1992 and 1997 Start practice 1992 study Eligible in study Response Response % 1977 ± % 1987 ± 1992 Ð Ð % 675 a % a 672 eligible.

3 Table 2 Predisposing factors, driving forces and perceived barriers of PCPs (the majority results from factor analysis), used in the LISREL-path analysis of `Noticing Patients' Overweight and Guidance of Treatment' S Description a a Item b a a Difference c P d Predisposing factors a) Interest in the effect of nutrition on health Ð 1 Ð a1 ˆ a2 NS b) Perception of own ability to in uence lifestyle and eating habits b2 < b1 P < of patients with health problems ( self-ef cacy factor) c) Perception of own ability to give dietary advice in the treatment c2 < c1 P < and prevention of coronary heart disease ( self-ef cacy factor) d) Perception of role of behaviour and heredity on health d1 ˆ d2 NS Driving forces e) Task perception e2 < e1 P < 0.01 f) Attitude regarding treatment of overweight f2 < f1 P < g) Attitude towards weight-health relationship Ð 1 Ð g1 ˆ g2 NS Perceived barriers h) Lack of skills to treat overweight h1 ˆ h2 NS i) Lack of time to treat overweight i2 < i1 P < 0.05 Dependent variable j) Noticing patients' overweight and guidance of treatment j2 < j1 P < a Crohnbach's alpha, a measure of reliability of the factors. b Number of items constituting a factor. c Indices 1 and 2 indicate the measurements of 1992 and 1997 (response). d P-value of paired t-test. demonstrated that nutrition guidance practices of PCPs are directly and signi cantly based on a few predisposing factors; driving forces and perceived barriers may act as signi cant intermediary variables. The predisposing factors, driving forces and perceived barriers were identi ed. The LISREL-model of mechanism of action of determinants of `noticing patients' overweight and guidance of treatment' for PCPs in practice from 5 up to 15 y in that publication (Hiddink et al, 1997b) will be referred to in this paper as the `reference-lisrel-model'. The nine factors used in the LISREL path analysis for analyzing the dependent variable `noticing patients' overweight and guidance of treatment' are given in Table 2. These factors were constructed in the same way for 1992 and Crohnbach's alpha (Cr-a) was used as a measurement of reliability of scales. Factors were de ned as sums of items, standardized for scale width. Of these nine factors, ve factors have a high Crohnbach's a ( > 0.70) and two have a moderate Crohnbach's a (0.60 ± 0.70). The other two factors were based on one question. The dependent variable as measured in 1997 was signi cantly lower compared to 1992 (P < 0.001). Both have also a moderate Crohnbach's a (0.66 and 0.62, see Table 2). Hypothesis Our hypothesis is in essence that determinants of the dependent variable (a nutrition guidance practice of primary care physicians (PCPs)), as well as their mechanism of action, are identical, both cross-sectionally and longitudinally, to the determinants of the reference LISRELmodel (and to their mechanism of action). In other words, we expect to obtain LISREL-models that t the empirical data, when we execute the LISREL path analysis with the predisposing factors, driving forces and perceived barriers of Table 2, for the dependent variable in different situations. These situations are: (a) Representative cross-sectional analysis: all respondents, in practice from 5 up to 20 y (N ˆ 371, see Table 1). (b) Cohort cross-sectional analysis: respondents in practice from 10 up to 20 y (N ˆ 283, see Table 1). (c) `Determinant-longitudinal approach' (with incorporation of the factor `time'). On the basis of the predisposing, enabling and reinforcing factors in the Precede- Proceed Model (Green & Kreuter, 1991) and the results obtained thus far (Hiddink et al, 1997b), we hypothesized that the dependent variable as measured in 1997 for the respondents in practice for between 10 and 20 y (N ˆ 283) is determined by the predisposing factors of 1992, the change in predisposing factors between 1997 and 1992 (D predisposing factors); the intermediary factors of both 1992 and 1997 may act as intermediary variables (see Figure 1). (d) `Early behaviour longitudinal approach' (with incorporation of the factor `time' and the factor `earlier behaviour'). It is well-known from the literature (De Vries et al, 1995) that earlier behaviour has a high predictive value for that behaviour to be carried out. Therefore, we now added in the hypothesis the variable `noticing patients' overweight and guidance of treatment' of 1992 as a driving force and hypothesized that the dependent variable as measured in 1997 for our respondents in practice for between 10 and 20 y (N ˆ 283) is determined by the predisposing factors of 1992, the change in predisposing factors between 1997 and 1992 (D predisposing factors), and `noticing patients' overweight and guidance of treatment' of 1992; the intermediary factors of 1992 and 1997 may act as intermediary variables (see Figure 2). Statistics Variables were constructed in the same way as in the previous study (Hiddink et al, 1997b). Associations were studied using Pearson's correlation test. Differences between groups were tested by t-test or the chi-square test. Differences over time were tested using paired t-test. To identify the mechanism of action of determinants of the dependent variable, linear structural relationships analysis

4 S38 Figure 1 Postulated model of mechanism of action of determinants of noticing patients' overweight and guidance of treatment for PCPs in practice for between 10 and 20 y (the `determinant longitudinal approach'). Figure 2 Postulated model of mechanism of action of determinants of noticing patients' overweight and guidance of treatment for PCPs in practice for between 10 and 20 y (the `early behaviour longitudinal approach').

5 (LISREL path analysis) was used (program version 8.14; JoÈreskog & SoÈrbom, 1989). The LISREL-solution was accepted, when the conditions of (a) low residuals, (b) all t-values of effects > 2 and (c) an acceptable Q-plot of all standardized residuals were ful lled (JoÈreskog and SoÈrbom, 1989). 2 DF, P, and adjusted goodness of t index (AGFI) are presented to indicate the quality of the model. Results Response and representatives The net response rate was 55% (371=672, see Table 1). The 371 respondents were well representative of the population of PCPs who had been in practice for 5 up to 20 y according to gender distribution, age, year of starting practice and type of practice. The respondents were also well representative according to sex by type of practice by health center (yes=no) distribution. The respondents did not differ from non-respondents according to gender distribution, age, year of starting practice, type of practice and degree of urbanization. Respondents and non-respondents were also compared on the basis of 22 variables of the 1992 study using t-test with Bonferroni-correction, including perceived barriers to nutrition guidance practices (6 factors) and 16 factors in the elds of nutrition attitudes and beliefs (Hiddink et al, 1997). This comparison revealed no statistical differences. Change of factors between 1997 and 1992 The change of predisposing factors, driving forces, perceived barriers and the dependent variable `noticing patients' overweight and guidance of treatment' of PCPs is given in Table 2. Between 1997 and 1992, two of the four predisposing factors, two of the three driving forces and one of the two perceived barriers decreased signi cantly. In those cases the factor did not change, the individual values for that variable may have changed. At the same time the dependent variable also decreased signi cantly (expressed in z-score, from z 1 ˆ 0:10 to z 2 ˆ 0:34). Representative cross-sectional analysis The hypothesis that the dependent variable is determined by predisposing factors, with or without intermediary factors (see Methods), could be con rmed because the LISREL-program also provided a model with an excellent t (Figure 3). The obtained LISREL-model of Figure 3 ts the empirical data ( 2 DFˆ24 ˆ 22:87, P ˆ 0.53; the AGFI ˆ 0.97). The percentage of explained variance in `noticing patients' overweight and guidance of treatment' by the LISREL-model is 23%. From this LISREL-model (Figure 3) it becomes clear that all predisposing factors act, in principle, though intermediary factors on the dependent variable. Only the predisposing factor `interest in the effect of nutrition in health and disease' acts directly on this dependent variable, the direct effect being the most important one. The model of Figure 3 very closely resembles the `reference-lisrel-model' with also an explained variance of 23%. Of the 15 paths in Figure 3, 12 paths are identical S39 Figure 3 LISREL-model of mechanism of action of determinants of noticing patients' overweight and guidance of treatment for PCPs in practice for between 5 and 20 y (the representative cross-sectional analysis).

6 S40 to paths in the reference model with almost the same path coef cients. Four paths in the `reference-lisrel-model' are not present in Figure 3 but replaced by three new paths. The conclusion may be, that the same set of predisposing factors and intermediary factors can explain the dependent variable for two different representative crosssectional studies of PCPs: being in practice for 5 up to 15 y in 1992 (`reference-lisrel-model') and being in practice for 5 up to 20 y in 1997 (Figure 3). Cohort-cross-sectional analysis The sub-sample of 283 respondents (in practice for 10 up to 20 y) who were also studied in 1992 were submitted to the same LISREL-analysis. The obtained LISREL-model of Figure 4 ts the empirical data: 2 DFˆ25 ˆ 25:17, P ˆ 0.45; AGFI ˆ 0.96). The percentage of explained variance by the LISREL-model is 21%. From this LISREL-model (Figure 4) it becomes clear that all predisposing factors act, in principle, through intermediary factors on the dependent variable. Only the predisposing factor `interest in the effect of nutrition in health and disease' acts directly on this dependent variable, the direct effect being the most important one. The model of Figure 4 very closely resembles the `reference-lisrel-model' with an explained variance of 23%. Thirteen paths are identical to the paths in the reference model, with almost the same path coef cients, and four paths in the `reference-lisrel-model' are not present in Figure 4 but replaced by one new path. The model of Figure 4 also very closely resembles the model of Figure 3. Of the 15 paths in the model of Figure 3, 13 paths are identical (with almost the same path coef cients), two paths are not present in Figure 4 and Figure 4 has one new path. The conclusion may be that the same set of predisposing factors and intermediary factors can explain the dependent variable in a cohort cross-sectional study at one point in time and ve years later. `Determinant-longitudinal approach' The hypothesis that the dependent variable as measured in 1997 for the respondents in practice for 10 ± 20 y (N ˆ 283) is determined by predisposing factors of 1992, the change in predisposing factors between 1997 and 1992 (D predisposing factors), with or without the intermediary factors of 1992 and of 1997 (Figure 1) could be con rmed because the LISREL-program also provided a model with a good t ( 2 DFˆ110 ˆ 131:44, P ˆ 0.080; AGFI ˆ 0.92). The percentage of explained variance is 21%. The conclusion may be, that incorporation of time as operationalized in this `determinant-longitudinal approach' makes it possible to obtain a LISREL-model of the dependent variable that ts the empirical data. The percentage of explained variance is the same as in the cohort crosssectional analysis (Figure 4, 21%), so the followed procedure did not lead to a gain in explained variance. `Early behaviour longitudinal approach' The hypothesis that the dependent variable as measured in 1997 for the respondents being in practice for 10 ± 20 y Figure 4 LISREL-model of mechanism of action of determinants of noticing patients' overweight and guidance of treatment for PCPs in practice for between 10 and 20 y (the cohort cross-sectional analysis).

7 (N ˆ 283) is determined by predisposing factors of 1992, the change in predisposing factors between 1997 and 1992 (D predisposing factors), and `noticing patients' overweight and guidance of treatment' of 1992, with or without the intermediary factors of 1992 and 1997 (Figure 2) could be con rmed, because the LISREL-program also provided a model with a good t ( 2 DFˆ123 ˆ 133:49, P ˆ 0.24; AGFI ˆ 0.92). The percentage of explained variance is 29%, which is higher than the explained variance of the `determinant-longitudinal approach' (21%) and also higher than the explained variance (24%) of the behaviour in 1997 from the behaviour in In this case, the followed procedure led to a gain in explained variance. Discussion This study recon rms that a PCPs nutrition guidance practice is determined partly directly by predisposing factors, and indirectly via driving forces and barriers which play an intermediary role. In the constructed LISREL-models (Figures 3 and 4), four predisposing factors were identi ed as playing a major role: ± perception of the physician's own ability to in uence the lifestyle and eating habits of patients with health problems (self-ef cacy general); ± physician's interest in the effect of nutrition in health and disease (nutritional interest); ± perception of physician's own ability to give dietary advice in the treatment and prevention of coronary heart disease (self-ef cacy chd); ± perception of role of behaviour and heredity on health (role of behaviour on health). These four predisposing factors also play a major role in the reference LISREL-model which con rmed a postulated model of action (Hiddink et al, 1997b). The arguments for that postulated model were based on the predisposing, enabling, and reinforcing factors in the Precede-Proceed Model (Green & Kreuter, 1991) and upon our previous qualitative research, which revealed these four predisposing factors for the dependent variable (Hiddink et al, 1997b). In the constructed LISREL-models (Figures 3 and 4) the same three driving forces (task perception, attitude regarding treatment of overweight, attitude towards weight-health relationship) and the same two perceived barriers (lack of skills to treat overweight, lack of time to treat overweight) act as intermediary variables, compared with the reference LISREL-model. The postulated hypotheses for the representative crosssectional analysis, the cohort cross-sectional analysis, the `determinant-longitudinal approach' and for the `early behaviour longitudinal approach' could be con rmed. Determinants of the examined nutrition guidance practice of PCPs, as well as their mechanism of action, are identical in these cross-sectional and longitudinal situations. LISREL-models clearly have advantages compared to the more traditional multiple regression analysis (MRA) in leading to a more in-depth understanding of the mechanism of action of determinants of nutrition guidance practices of PCPs (Hiddink et al, 1997b). MRA delivers beta-weights regarding the strength of the effect, but not an understanding of the mechanism of action (Hiddink et al, 1997b). A better understanding of this mechanism of action could be extremely bene cial when planning interventions to improve the nutrition guidance practices of PCPs (Green & Kreuter, 1991). Both in the representative cross-sectional analysis (Figure 3) and in the cohort cross-sectional analysis (Figure 4) the hypothesis that the dependent variable is determined by predisposing factors, for which driving forces and perceived barriers may act as intermediary variables, could be con rmed because the LISREL-program provided a model with an excellent t. The same set of predisposing factors and intermediary factors can partly explain the dependent variable (a) for two different representative cross-sectional studies of PCPs (being in practice for 5 to 15 y in 1992 (the `reference- LISREL-model') and being in practice for 5 to 20 y in 1997 (Figure 3)) (b) in a cohort cross-sectional study at one point in time and 5 y later. In all situations the mechanism of action of determinants is identical. In the constructed LISREL-models of the `determinantlongitudinal approach' and of the `early behaviour longitudinal approach', the four earlier mentioned predisposing factors are playing a major role, as well as three of the four `D predisposing factors'. Comparison of these two LISREL-models reveals the following features: ± in the `determinant-longitudinal approach' there are no direct effects of predisposing factors on the dependent variable, except for `nutritional interest', and `D nutritional interest'. ± the structure among determinants is almost identical: the `determinant-longitudinal approach' has 33 paths and the `early behaviour longitudinal approach' has 36 paths; 28 paths are identical (with almost the same path coef cients). Both in the `determinant-longitudinal approach' and in the `early behaviour longitudinal approach' the postulated hypothesis (see hypotheses of Figures 1 and 2) could be con rmed because the LISREL-program provided a model with a good t. Incorporation of time in the LISREL-model as operationalized in the `determinant-longitudinal approach' had no bene t with regard to the percentage of explained variance (21%) compared with the LISRELmodel of the cohort cross-sectional study (21%). Incorporation of time and earlier behaviour as operationalized in the `early behaviour longitudinal approach' led to a higher percentage of explained variance (29%) than in the cohort cross-sectional study (21%) and than in the `determinantlongitudinal approach' (21%). On the basis of the earlier behaviour alone the explained variance was 24%. The conclusion may be that incorporation of time and earlier behaviour as operationalized in the `early behaviour longitudinal approach' provided a model with a good t and with the highest percentage of explained variance. This longitudinal study shows that the dependent variable `noticing patients' overweight and guidance of treatment' decreased signi cantly in 5 y time (P < 0.001, see Table 2). We therefore conclude that there is an urgent need for a theory-based, planned nutrition education of PCPs to counteract this trend. In the Netherlands, the prevalence of overweight and obesity de ned by body mass index (BMI) of 25 ± 30 kg=m 2 and 30 kg=m 2, respectively, is 30 ± 40% (Mathus-Vliegen, 1998a). Based on the projects `Monitoring risk factors for health in the Netherlands (MORGEN; 20 ± 59 y) and `Erasmus Rotterdam health of elderly research' (ERGO, above 55 y), 7% of the men and 11% S41

8 S42 of the women are obese (Kramers & Ruwaard, 1997). Obesity is becoming a global epidemic (WHO, 1998); also in the Netherlands overweight and obesity are a growing problem for all ages (Mathus-Vliegen, 1998a). In the Netherlands, coronary heart death is to be attributed 15 ± 30% to overweight and 4 ± 25% to obesity. For diabetes mellitus these percentages are 64 ± 82% and 33 ± 75%, respectively (Mathus-Vliegen, 1998b). Minor weight loss (5 ± 10% of body weight) in obesity appears to reduce the risk (Pijl & Meinders, 1998). On the basis of the unique position of the PCP as a source of nutrition information (high referral scores, high perceived expertise and reach to nearly all segments of the population (Hiddink et al, 1997c) and the potential for individual counselling in family practice (Van Weel, 1997; Van Binsbergen, 1997), PCPs have a unique potential role in prevention and treatment of overweight and obesity (a growing problem for all ages in the Netherlands) within a multi-faceted approach. This multi-faceted approach is described by the WHO-report (1998) and by Mathus-Vliegen (1998b). PCPs are familiar with the Body Mass Index (Hiddink et al, 1997d), which is encouraging because prevention and treatment of overweight and obesity starts with a valid assessment. Our results show that in 5 y time also two of the four predisposing factors and two of the three driving forces decreased signi cantly. A nutrition education policy aimed at PCPs and the training of PCPs should include at the same time a strengthening of the predisposing factors, strengthening of the driving forces and reduction of the barriers to reach the best possible outcome (Hiddink, Thesis 1996). One of the predisposing factors is the interest of PCPs in the effect of nutrition on health. Such interest is the second essential step after the factor awareness in the process of diffusion of innovations (Rogers, 1983). Without interest, further steps in this process are impossible. Other predisposing factors identi ed are the factor `in uence of behaviour on health', and the self-ef cacy variables `perception of own ability to in uence lifestyle and eating habits of patients with health problems' and `perception of own ability to give dietary advice in the treatment and prevention of coronary heart disease'. The notion of self-ef cacy is a component of Bandura's social cognitive theory, which attempts to explain the determinants of behaviour and its modi cation (Strecker et al, 1986; Bandura, 1977). A number of strategies outlined in Bandura's social cognitive theory can be of help to increase PCP's self-ef cacy: these are verbal persuasion, vicarious experience or modeling and personal experience (Bandura, 1986). In case of verbal persuasion it is necessary to present data showing that PCPs can effectively help patients to change their eating pattern and lifestyle. In the case of vicarious experience or modeling, PCPs need to be exposed to peers who feel con dent and successful in helping patients to change their eating pattern and lifestyle. In the case of personal experience, a structured dietary treatment program on the basis of sound behaviour change principles can help to guide PCP's counselling, which increases the likelihood of successful dietary change among patients. Clearly, a future research priority is con rmation of these results for other nutrition guidance practices of PCPs as well as for PCPs-in-vocational-training. Together with the present results, this will lead to a better understanding of the determinants of nutrition guidance practices of PCPs (and their mechanism of action), which can be very bene cial when planning interventions to improve the nutrition guidance practice of PCPs (Green & Kreuter, 1991). With the results of our studies and the knowledge of strategies to increase PCP's self-ef cacy, a next necessary step in future research will be an intervention study to prove that increasing the self-ef cacy of PCPs will lead to an improvement of nutrition guidance practices of PCPs. Recently, a major review was published on the effectiveness of nutrition education and implications for nutrition education policy, programs and research (Contento et al, 1995). With regard to health professionals it concluded that effective and successful programs are behaviorally focused, based upon appropriate theory and prior research, and include motivational messages in educational strategies (Contento et al, 1995). A very strong motivational message would be the success of the proposed intervention study. Acknowledgements ÐSupported by research grants from the Dairy Foundation on Nutrition and Health, Maarssen, the Netherlands. References Ammerman AS, DeVellis RF, Carey TS, Keyserling TC, Strogatz DS, Haines PS, Simpson RJ & Siscovick DS (1993): Physician-based diet counselling for cholesterol reduction: current practices, determinants and strategies for improvement. Prev. Med. 22, 96 ± 109. Bandura A (1977): Self-ef cacy: Toward a unifying theory of behavioural change. Psychol. Rev. 84, 191 ± 215. Bandura A (1986): Social Foundations of Thought and Action: A Social Cognitive Theory. New York, NY: Prentice-Hall. Van Binsbergen JJ (1997): Nutritional factors in Dutch family medicine: an inventory. Am. J. Clin. Nutr. (Suppl) 65, S1967 ± S1973. Contento I, Balch GI, Bronner YL, Lytle LA, Maloney SK, Olson CM & Sharaga Swadener S (1995): The effectiveness of nutrition education and implications for nutrition education policy, programs and research: a review of research. J. Nutr. Educ. 27, 277 ± 420. Dillman DA (1978): Mail and Telephone Surveys. The Total Design Method. New York: John Wiley. van Dusseldorp M, Meeuws H, van Kessel H, Hendriks L, Chin L & Bakx C (1988): Frequentie van voedingsvragen op het spreekuur van de huisarts. Ned. Tijdschrift. Geneeskd. 132, 2325 ± Glanz K & Gilboy MB (1992): Physicians, preventive care and applied nutrition: selected literature. Acad. Med. 67, 776 ± 781. Glanz K (1997): Review of nutritional attitudes and counselling practices of primary care physicians. Am. J. Clin. Nutr. (suppl) 65, S2016 ± S2019. Glanz K, Tziraki C, Albright CL & Fernandes J (1995): Nutrition assessment and counselling practices: attitudes and interests of primary-care physicians. J. Gen. Intern. Med. 10, 89 ± 92. Green LW & Kreuter MW (1991): Health Promotion Planning. An Educational and Environmental Approach, 2nd edition, London: May- eld Publishing Company. Hiddink GJ (1992): Voedingsvoorlichting voor en door huisarten. In: Van mond tot mond; voorlichting over voeding. Series Nutrition and Health, part 24 (Stasse Wolthuis M. & Kok GJ, Eds) Bohn Sta eu van Loghum, Houten=Zaventem: 58 ± 71. Hiddink GJ (1996): Determinants of Nutrition Guidance Practices of Primary Care Physicians. PhD Thesis, Wageningen Agricultural University, The Netherlands. Hiddink GJ, Hautvast JGAJ, van Woerkum CMJ, Fieren CJ & van't Hof MA (1995): Nutrition guidance by primary-care physicians; perceived barriers and low involvement. Eur. J. Clin. Nutr. 49, 842 ± 851. Hiddink GJ, Hautvast JGAJ, van Woerkum CMJ, Fieren CJ & van't Hof MA (1997a): Driving forces for and barriers to nutrition guidance practices of Dutch primary-care physicians. J. Nutr. Educ. 29, 36 ± 41. Hiddink GJ, Hautvast JGAJ, van Woerkum CMJ, Fieren CJ & van't Hof MA (1997b): Nutrition guidance by primary-care physicians: LISRELanalysis improves understanding. Prev. Med. 26, 29 ± 36. Hiddink GJ, Hautvast JGAJ, van Woerkum CMJ, Fieren CJ & van't Hof MA (1997c): Consumers' expectations about nutrition guidance: the importance of primary-care physicians. Am. J. Clin. Nutr. (Suppl) 65, S1974 ± S1979.

9 Hiddink GJ, Hautvast JGAJ, van Woerkum CMJ, Fieren CJ & van't Hof MA (1997d): Information sources and strategies of nutrition guidance used by primary-care physicians. Am. J. Clin. Nutr. (Suppl) 65, S1996 ± S2003. JoÈreskog KG & SoÈrbom D (1989): LISREL VIII: A Guide to the Program and Applications, 2nd edition. Chicago: SPSS, Inc. Kramers PGN & Ruwaard D (1997): Eds Volksgezondheid Toekomst Verkenning Amsterdam: Elsevier=De Tijdstroom. Kushner RF (1995): Barriers to providing nutrition counselling by physicians: a survey of primary-care physicians. Prev. Med. 24, 546 ± 552. Levine BS, Wigren MM, Chapman DS, Kerner JF, Bergman RL & Rivlin RS (1993): A national survey of attitudes and practices of primary-care physicians relating to nutrition: strategies for enhancing the use of clinical nutrition in medical practice. Am. J. Clin. Nutr. 57, 115 ± 119. Mann KV & Putnam RW (1989): Physicians' perception of their role in cardiovascular risk reduction. Prev. Med. 18, 45 ± 58. Mathus-Vliegen EHM (1988a): Overgewicht I. Prevalenties en trends. Nederlands Tijdschrift voor Geneeskunde 142, 1982 ± Mathus-Vliegen EHM (1998b): Overgewicht II. Determinanten van overgewicht en strategieeèn voor preventie. Nederlands Tijdschrift voor Geneeskunde 142, 1989 ± Murphy P & Vogel E (1984): Nutrition education in a family practice residency program. Can. Fam. Phys. 30, 1646 ± `Nutritional attitudes and practices of primary care physicians'. Guesteditor Truswell AS. Proceedings of a symposium held in Heelsum, December 11 ± 13, Am. J. Clin. Nutr. 1997: 65 (Suppl), S1927 ± S2022. Report of a WHO consultation (1998): Obesity: Prevention and Managing the Global Epidemic. Geneva: WHO. Orleans CT, George LK, Houpt JL & Brodie KH (1985): Health promotion in primary care: a survey of U.S. family practitioners. Prev. Med. 14, 636 ± 647. Pijl H & Meinders AE (1998): Overgewicht III. Gewichtsreducerende behandelingsstrategieeèn. Nederlands Tijdschrift voor Geneeskunde 142, 1995 ± Rogers EM (1983): Diffusion of Innovations. New York: The Free Press. SPSS Reference Guide (1990): Chicago: SPSS, Inc. Strecker VJ, De Vellis BM, Becker MH & Rosenstock IM (1986): The role of self-ef cacy in achieving health behaviour change. Health Edu. Q 13, 73 ± 91. Thompson SC, Schwankovsky L & Pitts J (1993): Counselling patients to make lifestyle changes: the role of physician self-ef cacy, training and beliefs about causes. Family Practice 10(1), 70 ± 75. De Vries H, Backbier E, Kok G & Dijkstra M (1995): The impact of social in uences in the context of attitude, self-ef cacy, intention, and previous behaviour as predictors of smoking onset. J. Appl. Soc. Psychol. 25(3), 237 ± 257. De Vries, Dijkstra M & Kuhlman P (1988): Self-ef cacy: the third factor besides attitude and subjective norm as a predictor of behavioural intentions. Health. Educ. Res. 3, 273 ± 282. Wechsler H, Levine S, Idelson RK, Schor EL & Coakley E (1996): The physician's role in health promotion revisited: a survey of primary care practitioners. N. Engl. J. Med. 334, 996 ± 998. Van Weel C (1997): Morbidity in family medicine: the potential for individual nutritional counselling, an analysis from the Nijmegen Continuous Morbidity Registration. Am. J. Clin. Nutr. (Suppl) 65, S1928 ± S1932. Wiesemann A (1997): Nutritional counselling in German general practices: a holistic approach. Am. J. Clin. Nutr. (Suppl) 65, S1957 ± S1962. Worsley A (1989): Perceived reliability of sources of health information. Health Education Research 4, 367 ± 376. Young EA, Weser E, McBride HM, Page CP & Little eld JH (1983): Development of core competencies in clinical nutrition. Am. J. Clin. Nutr. 38, 800 ± 810. S43