Ethics statement

The Institutional Review Board of the Academic Medical Center (Medical Research Ethics Committee (MREC), Academic Medical Center, Amsterdam) approved the study. All participants provided oral and written informed consent.

Study population

All those studied were participants of the DHIAAN study, a randomized controlled trial (trial number NTR1499) in general practice. The aim was to study the effectiveness of a culturally targeted, intensive, lifestyle intervention to prevent type 2 diabetes and cardiovascular risk factors among Hindustani Surinamese The term ‘Hindustani-Surinamese’ refers to people of South Asian ancestral origin and their offspring who migrated to the Netherlands via Suriname. The Hindustani Surinamese are the descendants of the labourers from North India – Uttar Pradesh, Uttaranchal, and West Bihar – who were indentured between 1873 and 1917. The two large migration waves of Hindustani Surinamese to the Netherlands were caused mainly by the political situation in Suriname. The first wave took place at the time of the independence of Suriname in 1975, and the second wave, at the time of Desi Bouterse's coup in February 1980 [18].

The detailed methods of this study and the process of adapting the lifestyle intervention for the socio-cultural and socio-cognitive determinants of Hindustani Surinamese have been described earlier [19], [20]. In brief, 2307 Hindustani Surinamese (18–60 years old) living in The Hague, Netherlands, were screened via general practices between 18 May 2009 and 11 October 2010. In order to achieve a high response rate, we used a culturally targeted, intensive recruitment strategy that had been evaluated and proven feasible in the pilot of the DHIAAN study [20]. General practitioners sent each potential participant an invitation with a reply card that could be returned if further contact was unwanted. Invitees who had not responded received a written reminder and were contacted by telephone.

We asked all screening participants to fill out a brief questionnaire, undergo a physical exam, and provide a fasting blood sample. The 968 participants who were invited and screened between 18 May 2009 and 18 April 2010 also took an oral glucose tolerance test (75 g).

Inclusion in the trial

We invited the screening participants to participate in the trial if they had an impaired fasting glucose (5.7–6.9 mmol/l), an impaired glucose tolerance (2-h post-load glucose of 7.8–10.9 mmol/l), a glycated haemoglobin level of 42–46 mmol/mol, and/or a value of 2.39 or more for the homeostasis model assessment of estimated insulin resistance (HOMA-IR) [19] We excluded anyone who was already involved in a lifestyle programme, was pregnant, was known to have type 2 diabetes or a chronic disease that made participation in the intervention impossible, and/or used drugs interfering with plasma glucose levels. Furthermore, we excluded participants with newly diagnosed type 2 diabetes (i.e. a fasting glucose ≥7.0 mmol/l, a 2-h post-load glucose ≥11.0 mmol/l, or a haemoglobin (Hb)-A1c level ≥48 mmol/mol) and referred them to regular clinical care. We used a computer-generated randomisation list (individual one-to-one randomisation) to randomly assign 536 people to either the intervention group or the control group [19]. In brief, we generated a randomization list using simple randomisation (unstratified, no blocking) functions, and placed sealed envelopes representing the order of allocation into a box for the recruiters. Prior to randomisation, potential trial participants were only informed that the study would compare two lifestyle interventions, simply named programme 1 or programme 2, without detailing the specific design of these interventions. Sealed envelopes were opened in order of placement for each person who consented to randomisation and participation in the trial. Participants were then informed of the procedures for the arm of the trial that they had been assigned to. The masking (de facto masking) of the intervention and control group was maintained during the trial.

Intervention group

We offered all participants in the intervention group a lifestyle intervention. The design of this intervention was in line with the design of the proven efficacious intervention used in the Study on lifestyle intervention and impaired glucose tolerance Maastricht (SLIM), which aimed to evaluate the effect on glucose tolerance in a European Dutch population [21]. We culturally adjusted the intervention to the Hindustani Surinamese population, as cultural adaptations likely promote the effectiveness of interventions among specific ethnic populations [22]. Both surface and deep structure adaptations were used to make the intervention attractive, appropriate, and ultimately more potentially effective in our population [20], [23].

We used motivational interviewing to base the culturally appropriate, intensive, lifestyle intervention on individual lifestyle counselling. The lifestyle counselling in our study consisted of six to eight sessions (determined per individual, according to the goals specified protocol) in the first 6 months, followed by two booster sessions in the next 6 months, and this counselling aimed at both diet and physical activity. The counsellors were trained dieticians who were familiar with the Hindustani Surinamese culture and dietary habits. We offered the participants a family session with the dietician to decrease the social pressure to eat unhealthily and to increase the social support for a healthful lifestyle within the family. If desired, participants could also attend two cooking classes to improve their skills in how to adjust traditional dishes to make them more healthful.

Furthermore, we offered to supervise a 20-week physical activity programme for all participants in the intervention group. This programme, “exercise on prescription”, has been described elsewhere [24], [25]. Trained coaches monitored the participation in the physical activity programme.

Control group

We invited the participants in the control groups to join two group sessions led by student dieticians (at baseline and after 1 year). The sessions provided generic information about type 2 diabetes and discussed current guidelines for diet and physical activity. These participants received two flyers (at 3 and 9 months) with simple generic lifestyle advice.

Data collection

One trial visit for the intervention group and one for the control group was planned at baseline and after 1 year [19]. The invitation procedures for these visits were similar to those for the intensive screening procedures. Participants who did not respond to the invitation for the follow up visit were contacted by telephone and received a written reminder. Besides a written confirmation of their appointment, all participants received a reminder text message on the day prior to their appointment.

A trained interviewer interviewed each participant face to face during the visit. The interviewer noted the self-identified ethnicity and determined the educational level. The categories of educational level were: low – secondary education, primary education, or less; middle – low vocational training, lower secondary education, intermediate vocational training, and higher secondary education; and high – higher vocational training or university. Trained research staff, who were blinded to the allocation of participants to the intervention or control groups, used a standardised protocol for physical examinations. They measured weight on a mechanical scale (Seca 761, Hamburg, Germany) to the nearest 500 g. Height, waist circumference, and hip circumference were measured to the nearest 0.01 m. Body fat was measured to the nearest 0.1% by means of bioelectrical impedance analysis (OMRON BF500, Hoofddorp, Netherlands). All anthropometric measurements were obtained twice and the means were used for analysis. Blood pressure was measured in the seated position (OMRON M5-1, Hoofddorp, Netherlands), up to five times. We calculated the mean from the first two measurements with less than 5 mm Hg difference in both systolic blood pressure and diastolic blood pressure [19].

Both at baseline and after 1 year, all participants provided a fasting blood sample and were offered an oral glucose tolerance test. Measurements of fasting plasma glucose (hexokinase, Roche Diagnostics [RD]), 2-h post-load glucose [75 gram] (hexokinase, [RD]), insulin (immunoassay, sandwich principle, [RD]), HbA1c (high-performance liquid chromatography), total cholesterol, high-density lipoprotein cholesterol, low-density lipoprotein cholesterol, and triglycerides (enzymatic colorimetric method for all cholesterol and triglyceride measurements, [RD]) were carried out according to a standardised protocol at the General Practice Laboratory Foundation, Etten-Leur, Netherlands. HOMA-IR was calculated as (fasting plasma glucose * fasting plasma insulin)/22.5 [26].

Main outcomes

At the time of registration of the trial, the intended primary outcome was the 3-year incidence of type 2 diabetes. However, as the initial response rate was lower than expected, the recruitment period had to be extended to 2 years. Due to a fixed end date of the study (grant restrictions), the follow-up time was reduced to 2 years. As this period was too short to properly investigate differences in incidence of type 2 diabetes between the control and intervention group, we changed the primary outcome to the more proximal outcomes, namely changes in weight and other weight-related measurements (body mass index, waist circumference, and fat mass) after 1 year. Secondary outcomes were changes in glucose metabolism, blood pressure, and lipid profile after 1 year. The changes of the primary outcomes of the trial were approved by the Ethics Committee of the Academic Medical Center [19].

Statistical analysis

In the current analysis, we included all those who participated both in the baseline measurement and the follow-up measurement after 1 year, but we excluded anyone without weight data (Figure 1). Of the 536 participants, 197 were lost to follow-up (of which 104 actively declined), and 4 did not have follow-up data available for the weight status. This left 335 participants (177 in the intervention group and 158 in the control group) with data available after a median follow-up time of 380 days (IQR 359–420 days). To evaluate potential selection, we compared baseline characteristics of those with follow-up data available after 1 year with characteristics of those who were lost to follow-up. Apart from being younger (43.3 years versus 45.1 years), those lost to follow-up and those who did have follow-up data after 1 year had similar baseline characteristics (Supplementary Material; Table S1). Furthermore, we found that the odds of being lost to follow-up were similar in the intervention and control group, both in a univariate analysis and in an analysis with adjustment for sex, age, baseline BMI and baseline HbA1c (adjusted OR for the intervention group versus the control group: 0.99 [95%CI 0.70–1.43]).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 1. Flow diagram.

Normoglycemia was defined as having a fasting glucose <5.7 mmol/l, a 2-h post-load glucose of <7.8, a glycated haemoglobin level of <42 mmol/mol, and a value <2.39 or more for the homeostasis model assessment of estimated insulin resistance. Those with known or newly diagnosed diabetes (i.e. a fasting glucose ≥7.0 mmol/l, a 2-h post-load glucose ≥11.0 mmol/l, and/or a haemoglobin (Hb)-A1c level ≥48 mmol/mol) were excluded from the study.

https://doi.org/10.1371/journal.pone.0068605.g001

We also determined whether those who remained in the intervention and control groups had similar baseline characteristics. We described the characteristics with means (with standard deviations) or percentages, and we tested for differences between groups using independent sample t-tests and chi-square tests.

Then we described the 1-year changes in weight status, other anthropometric measurements, blood pressure, lipid profile, and glucose metabolism. We used paired t-tests to compare these outcomes at baseline and after 1 year in both the intervention group and the control group. In order to determine the effectiveness of the intervention, we compared the changes in these measurements between those assigned to the intervention group and the control group at baseline. We used independent sample t-tests to study the differences in continuous measures and chi-square tests to compare the differences in categorical measures. We verified the consistency of the findings across age and sex groups. Additionally, we repeated the main analysis comparing the changes in the intervention versus the changes in the control group with non-parametric tests.

We carried out several additional analyses. First, to determine whether potential dependencies influenced our results between participants belonging to the same general practice, we used a two-level linear regression model in which participants (level 1) were nested within general practices (level 2). To allow for dependencies, we incorporated a random intercept (level 2) into the model. We then determined whether the estimates for the fixed effects in these analyses were similar to the results of the original analyses, and we graphically investigated the random components of the model. Given that there were only 14 people with family members in the study who had follow-up data available, we refrained from carrying out a similar multilevel analysis with the family data.

Moreover, in order to verify the robustness of our results after the loss to follow-up, we repeated the main analyses after using multiple imputation to replace the missing values for the changes after 1 year in weight status, blood pressure, lipid profile, and glucose metabolism [27]. In these analyses, all participants were included, as opposed to only those who had follow-up data available. The imputation involved creating five different datasets in SPSS 19.0 using Fully Conditional Specification procedures (with logistic regression for categorical variables and linear regression for continuous variables). All available data were included in the imputation model. Thereafter, the multiple datasets were analysed as described above, and pooled estimates were computed. We then determined whether the results of the analyses with the multiply imputed data differed from our results of the analyses with complete-case data.

Finally, we conducted several explanatory analyses to further examine potential factors associated with the effects of the intervention after 1 year within the intervention group. First, we described the participation in the different components of the intervention. Second, we compared differences in the change in weight status and the metabolic profile between those attending lifestyle counselling (≥2 sessions attended on top of an intake visit) and those not attending, and between those maintaining participation (≥6 sessions attended on top of an intake visit) and those not maintaining participation. We also distinguished between those who had participated in at least 2 components or in all components of the intervention with those who had not. We based these categories on the goals specified per session in the protocol of the intervention. [19].

We used SPSS 19.0 (SPSS, Chicago, Illinois) for all analyses.

Power calculation

The original power calculation has already been described [19]. In addition to the power calculation, we performed a post-hoc power calculation study to demonstrate differences in the main outcome (i.e. weight loss) using the numbers of participants who participated both in the baseline measurement and the follow-up measurement after 1 year and had complete data on weight status. As already mentioned, 177 participants in the intervention group and 158 participants in the control group had complete data available both at baseline and after 1 year. Previous studies have found a difference in weight reduction of 1.5–3 kg (SD of 4) after 1 year [12], [28]. Assuming an alpha of 5%, we had a power of 81% to demonstrate a minimum difference in weight reduction of 1.25 kg.

Main findings

We investigated the effectiveness of an intensive, culturally targeted lifestyle intervention in general practice among 18 to 60-year-old Hindustani Surinamese, a population of South Asian origin with a high risk of type 2 diabetes, living in the Netherlands. We found that the lifestyle intervention did not effectively change the weight status, glucose metabolism, blood pressure, or lipid profile in the target population. After 1 year of follow-up, the metabolic risk factors of the intervention and control groups were similar. The attendance and maintenance of the intervention did not explain the lack of change in weight and metabolic profile in the intervention group.

Limitations of the study

This study has both strengths and limitations that we wish to mention before discussing our main findings. One strength of the study is that it is unique in terms of population (South Asians in industrialised countries), age (young population), and the culturally targeted design of the intervention [20]. Furthermore, no studies have reported investigating the potential effectiveness of an intensive lifestyle intervention in practice in this population, while South Asians in industrialised countries are at high risk of type 2 diabetes and are therefore an important target group.

In spite of the intensive invitation procedures, there was a high loss to follow-up (197 of the 536 participants). This drop-out emphasizes the difficulty of realising health gain by means of a lifestyle intervention in general practice. At the same time, we consider this a limitation of our study. Although those lost to follow-up and those with follow-up data available after 1 year had similar baseline characteristics and loss to follow-up was not associated with group allocation, we cannot rule out selection bias in our study. We attempted to account for the effects of the loss to follow-up by repeating our analyses in a multiply imputed dataset. These analyses did not change our interpretation of the results.

Another limitation was that we measured weight to the nearest 500 g, which influenced the precision of the reported weight and weight change after 1 year. However, this applies equally to the measurements taken in the intervention group and the control group, so that it is unlikely that this limitation has majorly influenced the reported difference in change of weight between the two groups.

Discussion of the main findings

We found that a culturally targeted intensive lifestyle intervention in general practice did not effectively change the metabolic risk factors among our Hindustani Surinamese study population after 1 year of follow-up. Our results are in line with results from studies that have investigated the effectiveness in clinical practice of lifestyle interventions for the prevention of type 2 diabetes among other ethnic populations [13]–[16]. Although the effects in clinical practice have been smaller than in the large efficacy trials, some clinical studies have reported changes in metabolic risk factors. For example, Vermunt et al., who studied the overall effect of a lifestyle intervention on type 2 diabetes risk reduction in Dutch primary care, found that the changes over time between those receiving the lifestyle intervention and those receiving conventional care were small, and most of them were not significant [13]. In addition, Saaristo et al., who implemented an intensive lifestyle intervention in primary health care among residents with a high risk of developing type 2 diabetes, showed a weight reduction of approximately 1 kilogram after 1 year [16].

The results of our study pinpoint that, despite the potential efficacy among South Asians [17], realising a health benefit by means of an intensive lifestyle intervention among a high risk South Asian population in general practice is extremely difficult. This was somewhat unexpected. As South Asians are known to have a high a-priori risk of developing type 2 diabetes and have a familial disposition toward type 2 diabetes [3], we expected a high level of awareness and motivation to make behavioural changes, especially during a lifestyle intervention trial. Indeed, participants in focus group interviews carried out before our study (unpublished) pointed out the necessity of such an intervention for this population. In addition, the intervention in our study was targeted extensively to the Hindustani-Surinamese population[20], and the dieticians who performed the individual lifestyle counselling were all trained according to recommended protocols and instructed to monitor the participation of all those included in the study. For these reasons, despite the modest results of previous intervention studies in clinical practice but given the culturally targeted study design and the ethnic background of our population, we find it remarkable that we found neither changes in metabolic profile over time within the control and intervention groups, nor a difference in change over time between these groups. Given this discrepancy, we believe that understanding the effects of motivational factors and behavioural changes in the light of the lack of effect of our intervention, as well as a detailed process evaluation of the experiences in practice, will be of the utmost importance.

We do wish to mention that, although the lifestyle intervention was not effective in our overall population, we did find that participants who achieved a relatively large weight loss after 1 year (i.e. those within the highest quartile of weight loss), regardless whether they had been assigned to the control group or the intervention group, showed more improvement in lipid profile and glucose metabolism than participants in the lowest quartile of weight loss (supplementary material; Table S2). This emphasises that, if high-risk individuals in this population lose weight, there is a potential health gain.

At the same time, the low initial response rate and laborious recruitment [19], the high drop–out rate, and the lack of effect of the lifestyle intervention on weight change and other metabolic parameters raise the question whether our strategy is the optimal approach to prevent type 2 diabetes in our target population. Apart from fighting already existing metabolic disturbances with lifestyle interventions (i.e. in a high-risk approach), we might achieve additional health gain at by also focusing on prevention strategies that tackle the high risk of type 2 diabetes among South Asians in an even earlier stage [29].

All in all, we found that an intensive, culturally targeted lifestyle intervention did not result in improvement of weight status and metabolic profile after 1 year for South Asian individuals with a high risk of developing type 2 diabetes. Our results show that realising a health benefit by means of an intensive lifestyle intervention among a high risk South Asian population in general practice is extremely difficult and that additional preventive strategies may have to be considered in order to achieve the intended health gain.