Effectiveness of interventions in preventing injuries in agriculture--a systematic review and meta-analysis.

OBJECTIVES
This study reviewed the effectiveness of interventions in preventing occupational injuries among workers in agriculture.


METHODS
Randomized controlled trials, controlled before-after studies, and interrupted time-series studies assessing interventions aimed at preventing injuries among workers in agriculture were considered. MEDLINE and five other databases were searched up to June 2006. Two authors independently assessed the eligibility of studies and the methodological quality of the ones included. Randomized controlled trials were combined in a meta-analysis. Interrupted time-series studies were reanalyzed to assess the immediate and progressive effect on injuries.


RESULTS
Five randomized controlled trials and three interrupted time-series studies met the inclusion criteria. Six studies evaluated educational interventions and financial incentives, and two studies evaluated the effect of legislation. Three randomized controlled trials on educational interventions with 4670 adult participants did not indicate any injury-reducing effect, with a rate ratio of 1.02 (95% confidence interval 0.87-1.20), nor did two randomized controlled trials among children (6895 participants). Financial incentives decreased the injury level immediately after the intervention in one interrupted time-series study. Banning endosulfan pesticide in Sri Lanka led to a significant decrease in the trend of poisonings over time. Legislation requiring rollover protective structures on all tractors in Sweden did not produce a reduction in injuries, but the same requirement for new tractors was associated with a decrease in fatal injuries.


CONCLUSIONS
The reviewed studies provided no evidence that educational interventions are effective in decreasing injury rates among agricultural workers. Financial incentives may be a better means of reducing injury rates. Banning highly toxic pesticides may be effective. Legislation on safety devices on tractors yielded contradictory results.

among the most hazardous industries, along with mining and construction (2). �gricultural injuries are well documented in industrialized countries, but less so in developing countries. �n estimated 1�0 000 farm workers are killed each year, and millions more are either seriously injured in workplace accidents or poisoned with pesticides and other agrichemicals. It is likely that under-reporting is common, and the actual numbers are even higher (3).
�lthough a wide range of interventions has been developed, their effectiveness is not well understood. The most recent review of 25 farm safety interventions published in 2000 found little evidence that farm safety programs have been effective (�). While some studies have been able to report at least temporary changes in knowledge, attitudes, and behavior, none showed a sustained decrease in injuries or illnesses. Therefore, we conducted a systematic review to update and summarize the evidence on the effectiveness of interventions in preventing occupational injuries among agricultural workers.

Inclusion criteria
Studies had to meet the following three criteria to be included in the review: (i) the study participants had to be workers in the agricultural industry, workers primarily engaged in growing crops and animal production, (ii) fatal or nonfatal injuries had to be reported as an outcome measure, and (iii) the study design had to be a randomized controlled trial, a cluster randomized controlled trial, or a controlled before-after study, or have had an interrupted time series.

Data search
Our search terms covered the inclusion criteria for �agricultural work", �injury and safety", and �study design". The detailed search strategies have been given elsewhere (5). The following six general databases were searched through June 2006: the Cochrane Central Register of Controlled Trials, the Cochrane Injuries Group's specialized register, MEDLINE (from 1966), EMB�SE (from 1988), �sychINFO (from 1983), and OSH�-ROM (including NIOSH�TIC and H�SELINE). In addition, we searched seven agriculture-specific databases and three websites. Studies in any language were considered for inclusion.

Study selection, data extraction and quality assessment
Two researchers (ML and RR) independently screened the obtained titles and abstracts for eligibility, extracted data using a standardized form, and assessed the quality of the studies that met the inclusion criteria. The researchers were excluded from assessing their own studies. �rticles in languages other than English were reviewed by a native speaker. The methodological quality of the randomized controlled trials and interrupted time-series studies was assessed according to Downs & Black (6) and Ramsay et al (�), respectively. Missing data were requested from authors and received in all but two cases (8,9).

Quantitative data analysis
The intervention effect of the randomized controlled trials and the cluster randomized controlled trials was recalculated if necessary as the ratio of injury rates per 100 person-years of the intervention and the control groups. Similar interventions addressing either adults or children were combined using the natural logarithms of the rate ratios and the generic inverse variance method as implemented in RevMan �.2 (The Cochrane Centre, The Cochrane Collaboration, Copenhagen, Denmark, 2002) and described in the Cochrane Handbook of Systematic Reviews of Interventions (10). To facilitate the interpretation, we converted the pooled effect size and its confidence intervals back into a pooled rate ratio.
In the case of the cluster randomized controlled trials in which the clustering effect was not taken into account (11), we calculated the �effective sample size" by dividing the original sample size with the design effect, as described in the Cochrane Handbook (10).
For the interrupted time-series studies, the outcomes were reanalyzed according to Ramsay et al (�,12) and �idanapathirana (13). These methods utilize a segmented time-series regression analysis to estimate the effect of an intervention while taking into account secular time trends and any autocorrelation between individual observations. These reanalyses were performed with Stata 9.2 for Windows (StataCorp L�, College Station, TX, US�). The reanalysis made it possible to estimate the change in level and the change in trend after the intervention (�). � change in level, an immediate intervention effect, was defined as the difference between the observed level at the first intervention time point and that predicted by the pre-intervention time trend. � change in trend, a sustained effect of the intervention, was defined as the difference between the post-and pre-intervention slopes. For similar comparisons, we performed a meta-analysis separately for the changes in level and for the changes in trend, as recommended by Ramsay et al (12), using the generic inverse variance method as implemented in RevMan �.2.
The Springfeldt interrupted time-series study (1�) included the following four different years in which legislation was introduced: 1959 (�), 1965 (B), 19�0 (C), and 1981 (D). The data were divided into the Lehtola et al following four different time series surrounding these interventions: 195�-196�, 1960-1969, 1966-19�5, and 19��-1990. To prevent including the same data twice in the meta-analysis, we combined the introduction of legislation related to new tractors [requiring rollover protective structures in 1959 (�) and safety cabins in 19�0 (C)] on one hand and that related to all tractors [requiring rollover protective structures in 1965 (B) and cabins in 1981(D)] on the other.

Study selection
The references retrieved from the six main databases (N=�822) and the topic-related databases and websites (N=�9�) totaled 8616. From these, 122 potential full articles were selected. During the process of reviewing these articles, 10 new references were identified from the reference lists, the total number of references therefore being 132 articles. From the 132 articles, � could not be retrieved after an intensive search, 3 were reports concerning the same study, and 11� did not meet the inclusion criteria. Consequently we included 8 studies in the review [3 randomized controlled trials (8,15,16), 2 cluster randomized controlled trials (9,11), and 3 interrupted time-series studies (1�, 1�, 18)]. �fter a recheck of the missing titles, it did not appear that we missed any important studies.

Study characteristics
Three of the aforementioned studies were from the United States (8,9,16), two came from Finland (11, 1�), and one came from each of the following countries: Denmark (15), Sweden (1�), and Sri Lanka (18) (table 1). �ll of the studies were published between 1993 and 2006, and the interventions were carried out during 1959-2003.
Two studies (8,9) examined injury prevention among children or adolescents. The rest dealt with injury prevention among adults, and one of these (11) had only male reindeer herders as its participants.
�ll of the randomized controlled trials included a combination of the following educational interventions (table 2): (i) personal education by occupational health and safety (OH�S) professionals (eg, about farm safety checks, health checks, or safety courses), (ii) personal education by non-OH�S professionals, including other participants (eg, farm visits or group discussions), (iii) written information (eg, booklets, guides, mailings, written reports, booster interventions), and (iv) financial incentives (eg, travel expenses reimbursed or money paid to a participating farm).
Lee et al (9) combined all four of the aforementioned elements. Two studies (15,16) combined three of the elements, one study (8) used two, and one (11) contained only one of the elements. Two studies had two intervention groups and a control group (9,11). We used the most extensive intervention in the meta-analysis.
The interrupted time-series study of Rautiainen (1�) was based solely on incentives, studying how insurance premium discounts affect injury claims. �nother interrupted time-series study (18) evaluated pesticide regulation banning endosulfan use to decrease fatal poisonings. Springfeldt (1�) evaluated the effect of regulations requiring safety devices (technical measures) on tractors.

Methodological quality of the included studies
None of the studies attempted to blind its study participants, since blinding is difficult to accomplish in these types of studies. Only one study (8) reported blinding those who measured the outcome. Other issues decreasing the internal validity scores included noncompliance, unclear randomization, and no adequate adjustment for confounding. The maximum and minimum internal validity scores of the randomized controlled trials were 8 and 5 points, respectively, out of 13 points (table 3).
The quality of the three interrupted time-series studies were rated as follows: one study (1�) received 5 out of the maximum 8 points, and the other two studies (1�, 18) received only 3 points. The most common problem was the use of inappropriate time-series techniques for the analysis or no statistical analysis at all.

Effectiveness of educational interventions on injuries
Randomized controlled trials � meta-analysis of three randomized controlled trials (11,15,16) aimed at reducing injuries among adults showed no evidence of an effect on injuries [rate ratio 1.02, 95% confidence interval (95% CI) 0.8�-1.20] (figure 1). The narrow 95% confidence interval indicated that a more positive or more negative outcome is not very likely for these kinds of educational interventions. The studies were statistically homogeneous.
The two randomized controlled trials (8,9) aimed at reducing injuries among children and adolescents did not show a significant effect either, with a rate ratio of 1.2� (95% CI 0.51-3.16) (figure 2). H�owever, heterogeneity was high (I 2 =91.8%), as one study had a significant effect in favor of the control group.

Interrupted time-series studies
In one interrupted time-series study (1�), there was evidence that incentives have an immediate injury-reducing

Legislation banning the use of endosulfan
In one interrupted time-series study (18), there was evidence that an endosulfan ban had a progressive effect on the reduction of fatal injuries by poisoning. There was an increasing pre-intervention injury rate over time, as indicated by the positive slope (table �). The immediate effect was also significantly positive, meaning that the number of injuries continued to increase right after the intervention (effect size 2.20, 95% CI 0.9�-3.�3).

Legislation requiring technical measures on tractors
Technical measures on new tractors. In the meta-analysis of the time series dealing with the introduction of legislation requiring rollover protective structures or safety cabins on new tractors, there was a significant immediate and progressive increase in all injuries, but also an immediate and progressive decrease in fatal injuries with respective effect sizes of -0.90 (95% CI -3.38-1.58) and -0.93 (95% CI -1.82 --0.03) (1�).
Technical measures for all tractors. In the meta-analysis of the time series of the introduction of legislation

Farms or farmers
Annual health screening by trained nurse with one-on-one discussion regarding specific health concerns and proper use of personal protective equipment; annual on-farm safety review by trained farm safety consultant (local farmer) with discussion of hazard removal and safe work methods; informational meetings and focus groups to discuss aspects of the program Other participants in the informational meetings and focus groups · USD 200 each year paid to participating intervention farmer a Two of the studies had two intervention groups (standard and enhanced group) and a control group. Only one of the intervention groups could be selected for the meta-analysis; the selected group was the one that was thought to have received a more extensive intervention.

Summary of the main conclusions
We found no evidence in the meta-analyses to suggest that educational interventions had an injury-reducing effect. In two separate single studies, the introduction of insurance discounts reduced the level, but not the trend of injuries, and a legislative ban of endosulfan improved the trend for pesticide poisonings, but it was

Strength and limitations
We included only studies with robust study designs, and we were able to combine the results of similar intervention studies in the meta-analyses. This procedure increased the power to detect the effect of these interventions. The inclusion of only the studies that measured injuries as an outcome is another strength of our review.
It increased the quality of the evidence because the link between knowledge and attitudes and injury outcomes has not been strongly established (19). We conducted a very sensitive search and are confident that all of the studies that met our inclusion criteria were identified. In addition we had all non-English language abstracts read and interpreted by a person with appropriate language skills.
We included interrupted time-series studies, which are difficult to interpret in regard to the effect of the intervention. For instance, regulation may have an immediate effect in some cases or a delayed effect in others if the intervention initiates gradual changes. In some of the reviewed studies, the authors drew conclusions regarding the intervention effects on the basis of visual observation only. By statistically analyzing both immediate and progressive outcomes, our review provides a more robust assessment of the intervention impact in interrupted time-series studies.
We found no evaluations of engineering interventions, except the Swedish intervention with rollover protective structures, which we considered under regulatory interventions. The unexpected direction of some of our findings in relation to such structures is inconsistent with the prevailing view that engineering interventions are generally considered the most effective. Improvement in the design of machines, environments, and systems is preferable to attempts to change attitudes and behavior (20). Furthermore, risk compensation, among other confounding factors, may occur as a result of safer design. Therefore, further evaluations of such engineering interventions are needed.
�ltogether 90% of the farming population live in �sia and Sub-Saharan �frica. H�owever, all but one of the studies we found was based on data from industrialized countries (18). Most of our conclusions therefore may not be generalizable to developing countries, as the settings are so different.

Interpretation
The educational interventions in the reviewed studies may have been expected to show some effect, as they were comprised of combinations of different elements and engaged the study participants in different ways. H�owever, the effect sizes were small and not statistically significant. Similarly, the meta-analysis showed no effect, and the 95% confidence interval was narrow, indicating good precision. Only the study of financial incentives showed an effect on injury rates (1�). H�owever, due to financial incentives, it is possible that farmers underreported injuries.
Our negative findings are consistent with the results of other studies evaluating the impact of educational interventions alone on injury outcomes (21). These findings indicate that educational interventions are not adequate to bring about change, unless combined with incentives such as financial benefits or legislative requirements.
�esticides constitute a serious health hazard to farmers, especially in developing countries. H�owever, our review found only one study that addressed acute poisoning risk. This time series study showed that banning a toxic pesticide had a favorable effect on poisoning fatalities in the long term, and it did not lead to an illegal and more dangerous use of pesticides.
The Swedish study on rollover protective structures (1�) is frequently cited as strong evidence for the effectiveness of such structures (22). �n explanation for our contradictory findings could be that, even if the legislative changes were introduced to start on a certain date, it appears there were no clear interruptive effects since the percentage of tractors with rollover protective structures increased gradually, without major peaks. By the end of the study period, nearly 100% of tractors had such structures. It is interesting to note that the fatalities decreased to near zero early, much before the percentage of rollover protective structures reached full compliance. This finding appears to differ from experience in the United States, where about 60% of all tractors are currently equipped with rollover protective structures, but overturn fatalities are still common, about 100 per year (23, 2�). �nother limitation of the study was the fact that there were only two time points before the first legislation on rollover protective structures came into force in 1959, and it was therefore difficult to evaluate that intervention. Yet this initial legislation may have been the most important, initiating the steady increase in the percentage of tractors with rollover protective structures and a decrease in injury rates, particularly in the early years of the observation period.

Quality of the evidence
It is important to note that at least some of the included studies were designed as a randomized controlled study, although it is often argued that such a design is difficult or impossible to apply in occupational health settings.

Lehtola et al
H�owever, it remains difficult to perform high-quality studies, as the blinding of participants and providers is virtually impossible in educational interventions. None of the included studies scored more than �0% of the possible score on the quality checklists. H�owever, compared with the quality of evidence included in previous reviews, there was notable improvement in the quality of the studies (�).

Other reviews
Other reviews of interventions to prevent childhood farm injuries concluded that there was insufficient evidence to draw firm conclusions on their effectiveness (25,26). Three other reviews on general farm safety interventions (�, 2�), and educational interventions (28) using less strict inclusion criteria than we used, also concluded that there was no evidence available. We were able to include several randomized controlled trials and interrupted time-series studies not included in the previous reviews. This difference enabled us to conduct meta-analyses and draw conclusions on educational interventions from a larger number of studies.
�nother review on interventions to reduce pesticide poisonings concluded that exposures had been reduced but that there was a lack of evidence about whether concomitant poisonings had decreased (29). Our review found only one pesticide-related study that evaluated legislative intervention.
In general, this review revealed that only a small number of high-quality studies have been conducted in this area, providing a limited evidence base from which to inform prevention programs. It is clear that continued significant efforts are needed to develop and evaluate farm injury interventions.

Implications for practice
This review found no evidence supporting the widespread use of educational interventions alone. H�owever, there is likely a place for educational components within multifaceted interventions. The use of financial incentives could be effective but should be studied further before more extensive implementation can be recommended. The banning of endosulfan lowered the rate of fatal pesticide poisonings in one study and should be considered for other countries without such legislation.

Implications for research
Randomized controlled trials are possible and feasible both at the individual level and at the farm level. More of these studies are needed for evaluating behavioral interventions and interventions to enhance the implementation of engineering interventions. Interrupted time-series studies using administrative databases are feasible for studying the effects of interventions, particularly those at the society level (including legislative changes). Further studies should address the expected impact of legislation on the time series of injury rates. Finally, studies are needed that address farm safety problems in developing countries.