Aligning text mining and machine learning algorithms with best practices for study selection in systematic literature reviews.

BACKGROUND: Despite existing research on text mining and machine learning for title and abstract screening, the role of machine learning within systematic literature reviews (SLRs) for health technology assessment (HTA) remains unclear given lack of extensive testing and of guidance from HTA agencies. We sought to address two knowledge gaps: to extend ML algorithms to provide a reason for exclusion-to align with current practices-and to determine optimal parameter settings for feature-set generation and ML algorithms. METHODS: We used abstract and full-text selection data from five large SLRs (n = 3089 to 12,769 abstracts) across a variety of disease areas. Each SLR was split into training and test sets. We developed a multi-step algorithm to categorize each citation into the following categories: included; excluded for each PICOS criterion; or unclassified. We used a bag-of-words approach for feature-set generation and compared machine learning algorithms using support vector machines (SVMs), naïve Bayes (NB), and bagged classification and regression trees (CART) for classification. We also compared alternative training set strategies: using full data versus downsampling (i.e., reducing excludes to balance includes/excludes because machine learning algorithms perform better with balanced data), and using inclusion/exclusion decisions from abstract versus full-text screening. Performance comparisons were in terms of specificity, sensitivity, accuracy, and matching the reason for exclusion. RESULTS: The best-fitting model (optimized sensitivity and specificity) was based on the SVM algorithm using training data based on full-text decisions, downsampling, and excluding words occurring fewer than five times. The sensitivity and specificity of this model ranged from 94 to 100%, and 54 to 89%, respectively, across the five SLRs. On average, 75% of excluded citations were excluded with a reason and 83% of these citations matched the reviewers' original reason for exclusion. Sensitivity significantly improved when both downsampling and abstract decisions were used. CONCLUSIONS: ML algorithms can improve the efficiency of the SLR process and the proposed algorithms could reduce the workload of a second reviewer by identifying exclusions with a relevant PICOS reason, thus aligning with HTA guidance. Downsampling can be used to improve study selection, and improvements using full-text exclusions have implications for a learn-as-you-go approach.

Emergent biomarker derived from next-generation sequencing to identify pain patients requiring uncommonly high opioid doses.

Next-generation sequencing (NGS) provides unrestricted access to the genome, but it produces 'big data' exceeding in amount and complexity the classical analytical approaches. We introduce a bioinformatics-based classifying biomarker that uses emergent properties in genetics to separate pain patients requiring extremely high opioid doses from controls. Following precisely calculated selection of the 34 most informative markers in the OPRM1, OPRK1, OPRD1 and SIGMAR1 genes, pattern of genotypes belonging to either patient group could be derived using a k-nearest neighbor (kNN) classifier that provided a diagnostic accuracy of 80.6±4%. This outperformed alternative classifiers such as reportedly functional opioid receptor gene variants or complex biomarkers obtained via multiple regression or decision tree analysis. The accumulation of several genetic variants with only minor functional influences may result in a qualitative consequence affecting complex phenotypes, pointing at emergent properties in genetics.

[Thoracic endometriosis with aspecific manifestation]. / Thoracale endometriose met een aspecifieke uiting.

BACKGROUND: Endometriosis is a common disease, especially in subfertile women. The most common location is in the pelvis, but extragenital locations are also possible. This far less common condition has been described in almost all tissues in the body. Symptoms occurring cyclically are characteristic of endometriosis. CASE DESCRIPTION: A 37-year-old woman was discovered by chance to have ascites and pleural effusion. She had no symptoms of this. Thoracoscopy showed an image consistent with thoracic endometriosis. After initial drug therapy was unsuccessful, surgical intervention was performed. CONCLUSION: Thoracic endometriosis is a rare disease, in which the cyclical nature of the symptoms often leads to correct diagnosis. Drug therapy is the preferred treatment for patients.

Network meta-analysis of longitudinal data using fractional polynomials.

Network meta-analysis of randomized controlled trials (RCTs) are often based on one treatment effect measure per study. However, many studies report data at multiple time points. Furthermore, not all studies measure the outcomes at the same time points. As an alternative to a network meta-analysis based on a synthesis of the results at one time point, a network meta-analysis method is presented that allows for the simultaneous analysis of outcomes at multiple time points. The development of outcomes over time of interventions compared in an RCT is modeled with fractional polynomials, and the differences between the parameters of these polynomials within a trial are synthesized across studies with a Bayesian network meta-analysis. The proposed models are illustrated with an analysis of RCTs evaluating interventions for osteoarthritis of the knee. Fixed and random effects second order fractional polynomials were applied to the case study. Network meta-analysis with models that represent the treatment effects in terms of several parameters using fractional polynomials can be considered a useful addition to models for network meta-analysis of repeated measures previously proposed. When RCTs report treatment effects at multiple follow-up times, these models can be used to synthesize the results even if reporting times differ across the studies.

TOXICITY OF SEVERAL FUNGICIDES FOR ORANGE WHEAT BLOSSOM MIDGE, SITODIPLOSIS MOSELLANA (GÉHIN) (DIPTERA: CECIDOMYIIDAE).

The orange wheat blossom midge, Sitadiplosis mosellano (Géhin) (Diptera: Cecidomyiidae), can cause severe losses in wheat grain yield and quality. This pest is known to be susceptible to many insecticides, but various field observations have suggested that some fungicides could also significantly affect S. mosellana. In order to confirm these field observations, the effect on adult midges of several fungicides commonly applied to wheat crops was investigated in the laboratory and in small plots in the field. In each experiment, the fungicides were compared with a positive (insecticide) and a negative control (water). Four fungicides were assessed in the laboratory, each with five doses based on basis of a tenfold dilution starting at the field-recommended dose. The mortality rate was evaluated after 24 hours and the lethal dose 50% (LD50) was determined for each product. In the field, six fungicides were tested at the recommended dose. The effect of each product was compared on the basis of the number of S. mosellana adults caught alive with an insect vacuum sampler (Vortis®) on the morning after the treatments. Both experiments showed a significant effect of several fungicides tested on S. mosellana adults. Chlorothalonil was not toxic for S. mosellana, but tebuconazole, fluxapyroxad and azoxystrobin all induced significant mortality rates.

Pest select database: a new tool to use selective pesticides for IPM.

Crop pests can be naturally controlled by a set of beneficial arthropods, as entomophagous predators or parasites, leading to a global reduction of pest occurrence and insecticide use. The selectivity of plant protection products for these beneficial arthropods is a key factor in the success of IPM strategies. An abundant scientific literature concerning this subject exists, but for the final pesticide users, the information is not always easily accessible or understood. The IOBC working group "Pesticides and Beneficial Organisms" (PBO) has developed a new database to compile all data concerning pesticide selectivity on beneficial arthropods, obtained with methods that followed the IOBC test standard characteristics. These standards have been developed since the 70's by the Working group Pesticide and Beneficial Organisms, in the context of IPM. This database includes a first set of data previously published, coming from the different Joint Pesticide Testing Programs, with the addition of results published later in the IOBC Bulletin. The second set of data has been extracted from the Public verse of the Draft Assessment Report (DAR), established by the different EU member' states in the context of the registration process at the European Level and available online on the EFSA website. As nearly all active ingredients have to be tested on at least 2 to 4 selected beneficial arthropod species with methods that followed the IOBC standard characteristics, these DAR represent an important source of information for pesticide users dealing with beneficial arthropods and selectivity, with about 50% of the actual records. This source was previously only exploited for registration purposes and not accessible to a large public, but a lot of results, mainly from the first tier testing, have been compiled and interpreted for IPM. The new database is available online on the area restricted to the IOBC-members (http:// www.iobc-wprs.org/restricted_member/toolbox.cfm). This database will be updated regularly on basis of the new IOBC publications, EFSA Draft Assessment Report and scientific publication results that followed IOBC standard characteristics.

Accounting for correlation in network meta-analysis with multi-arm trials.

Multi-arm trials (trials with more than two arms) are particularly valuable forms of evidence for network meta-analysis (NMA). Trial results are available either as arm-level summaries, where effect measures are reported for each arm, or as contrast-level summaries, where the differences in effect between arms compare with the control arm chosen for the trial. We show that likelihood-based inference in both contrast-level and arm-level formats is identical if there are only two-arm trials, but that if there are multi-arm trials, results from the contrast-level format will be incorrect unless correlations are accounted for in the likelihood. We review Bayesian and frequentist software for NMA with multi-arm trials that can account for this correlation and give an illustrative example of the difference in estimates that can be introduced if the correlations are not incorporated. We discuss methods of imputing correlations when they cannot be derived from the reported results and urge trialists to report the standard error for the control arm even if only contrast-level summaries are reported. Copyright © 2012 John Wiley & Sons, Ltd.

A systematic review of the quality of homeopathic pathogenetic trials published from 1945 to 1995.

BACKGROUND: The quality of information gathered from homeopathic pathogenetic trials (HPTs), also known as 'provings', is fundamental to homeopathy. We systematically reviewed HPTs published in six languages (English, German, Spanish, French, Portuguese and Dutch) from 1945 to 1995, to assess their quality in terms of the validity of the information they provide. METHODS: The literature was comprehensively searched, only published reports of HPTs were included. Information was extracted by two reviewers per trial using a form with 87 items. Information on: medicines, volunteers, ethical aspects, blinding, randomization, use of placebo, adverse effects, assessments, presentation of data and number of claimed findings were recorded. Methodological quality was assessed by an index including indicators of internal and external validity, personal judgement and comments of reviewers for each study. RESULTS: 156 HPTs on 143 medicines, involving 2815 volunteers, produced 20,538 pathogenetic effects (median 6.5 per volunteer). There was wide variation in methods and results. Sample size (median 15, range 1-103) and trial duration (mean 34 days) were very variable. Most studies had design flaws, particularly absence of proper randomization, blinding, placebo control and criteria for analysis of outcomes. Mean methodological score was 5.6 (range 4-16). More symptoms were reported from HPTs of poor quality than from better ones. In 56% of trials volunteers took placebo. Pathogenetic effects were claimed in 98% of publications. On average about 84% of volunteers receiving active treatment developed symptoms. The quality of reports was in general poor, and much important information was not available. CONCLUSIONS: The HPTs were generally of low methodological quality. There is a high incidence of pathogenetic effects in publications and volunteers but this could be attributable to design flaws. Homeopathic medicines, tested in HPTs, appear safe. The central question of whether homeopathic medicines in high dilutions can provoke effects in healthy volunteers has not yet been definitively answered, because of methodological weaknesses of the reports. Improvement of the method and reporting of results of HPTs are required. REFERENCES: References to all included RCTs are available on-line at.

Pesticides selectivity list to beneficial arthropods in four field vegetable crops.

Selectivity of pesticides to beneficial arthropods is a key data for the implementation of IPM program. In the context of field vegetables crop, a set of 16 fungicides, 17 herbicides and 14 insecticides commonly used in Belgium were tested on 5 indicator species: the parasitic hymenoptera Aphidius rhopalosiphi (De Stefani-Perez) (Hym., Aphidiidae), the aphid foliage dwelling predators Adalia bipunctata (L.) (Col., Coccinellidae) and Episyrphus balteatus (Dipt., Syrphidae) and the ground-dwelling predators Aleochara bilineata (Col., Staphyllinidae) and Bembidion lampros (Col., Carabidae). Pesticides were tested according a testing scheme including a first assessment on inert substrate (glass plates for adults of A. rhopalosiphi, larvae of A. bipunctata and E. balteatus, sand on adults of A. bilineata and B. lampros) and, for product that were toxic, a second assessment on natural substrate (barley seedlings for A. rhopalosiphi, french bean plants for A. bipunctato and E. balteatus and two type of soil for 8. lampros and A. bilineato). The effects of the product were assessed on basis on mortality, except for A. bilineata (Onion fly pupae parasitism). According to the final results obtained at the end of this testing scheme, the product were listed in toxicity class: green list if effect < or =30%, yellow list 30% < effect < 60% and orange list 60% < effect < or =80%. Products with toxicity higher than 80% on plants or on soils, or that reduce parasitism more than 80% on soil were put in red list and are not recommended for IPM. Results showed that all fungicides and herbicides were included in the green list except tebuconazole and boscalid + pyraclostrobin that were labeled as yellow for A. bipunctata. In opposite, no foliar insecticide was totally selective for all beneficial tested. However some products are in green list for one or several species. Soil insecticides were all are very toxic for ground dwelling arthropods and classed in red list. All results obtained during this study and further upgrade will be available on www.cra.wallonie.be/selectivite. In conclusions, fungicides and herbicides tested are compatible with IPM programs. For foliar insecticides, some treatments can be used carefully according to the selectivity. But for soil insecticide treatments, their toxicity raise the question of their use in IPM programs in vegetables and the need of new compounds or development of alternative pest control programs.

Influence of organic matter on bio-availability of carbosulfan and its toxicity on a carabid beetle.

Study of factors influencing soil insecticide toxicity are needed to reduce negative impacts of these products on beneficial insects. To date, if high toxicity differences between different type of soils have been reported, there is few specific studies on soil parameters influence on selectivity of soil insecticides to beneficial arthropods. To assess the specific impact of organic matter, the relationship between bio-availability of a soil insecticide, carbosulfan [Sheriff 1 Gr], and its toxicity on a small Carabidae, Bembidion lampros (Herbst.) on a sand enriched with increasing quantities of organic matter was studied. In laboratory, adults of B. lampros were put on different substrate, made of sand or sand with addition of organic matter at 3, 6 and 9% w/w, and treated with carbosulfan applied as granule at the rate of 312.5, 625, 1250 and 6250 microg a.i./m corresponding respectively to 0.5, 1, 2 and 10% of the recommended field rate. Mortalities of B. lampros were assessed after 14 day of exposure. In parallel, the total carbosulfan residue (total extraction) and bioavailable fraction (CaCL2 aqueous extraction) were determined 48h after substrate treatments. According to the mortalities and bio-availability obtained, a dose--response relationship was calculated and compared with a reference relation dose--response obtained on sand, where the bio-availability of the product was considered as 100% of the amount of product applied. Carbosulfan was highly toxic on sand for B. lampros, with 100, 57 and 50% mortality at 10, 2 and 1% of the recommended field rate. When organic matter was added to the sand, the toxicity gradually decreased. This reduction in toxicity was more rapidly observed on sand + organic matter than on pure sand. The mortalities were strongly correlated with the bioavailability, indicating first that the organic matter is fixing an important part of the insecticide and secondly reduce its toxicity to beneficial arthropods. The results suggest that it could be possible, with further research, to predict toxicity of products in the field on different kind of soil according previous laboratory toxicity models and soil analysis. In conclusion, the organic matter influences strongly bio-availability of carbosulfan. This bioavailability was strongly correlated to toxicity to B. lampros. With the improvement of bioavailability determination and method validation, the assessment of pesticide bio-availability in the substrate could help to estimate the pesticide toxicity towards carabidae on different type of soils.

Predicting the long term course of low back pain and its consequences for sickness absence and associated work disability.

BACKGROUND: Low back pain is characterised by a dynamic pattern of episodes and recovery but little is known about the long term course of back pain due to lack of cohort studies with sufficiently long follow up periods. METHODS: A cohort of 523 workers in nursing homes and homes for the elderly was followed for two years. Physical load was measured by observations at the workplace. Psychosocial factors at work, individual characteristics, and low back pain were determined by questionnaire once a year. The effect of work load on low back pain and the transition of low back pain into sickness absence was calculated with logistic regression analysis. A Markov model was used to construct a hypothetical cohort of workers with follow up of 40 years (40 cycles of 1 year) with transitional probabilities between no complaints, low back pain, and sickness due to low back pain. Permanent disability was used as end state of health. RESULTS: The transitional probability from no complaints to low back pain varied between p = 0.25 and p = 0.29, from low back pain to sickness absence between p = 0.09 and p = 0.25, and recurrence of sickness absence varied between p = 0.27 and p = 0.50, depending on the level of physical load. During a 40 year career, total sickness absence due to low back pain was approximately 140 weeks (6.6%) among workers with high physical load and about 30 weeks (1.4%) among those with low physical load. CONCLUSION: The Markov approach illustrated the potential impact of physical load on (permanent) disability due to low back pain among workers with exposure to physical load. These consequences may go unnoticed in cohort studies with follow up periods of a few years.

Economic evaluation of voriconazole in the treatment of invasive aspergillosis in the Netherlands.

OBJECTIVE: To asses the cost-effectiveness of voriconazole in comparison to conventional amphotericin B and itraconazole for the treatment of invasive aspergillosis in the Netherlands. METHODS: The cost-effectiveness of voriconazole in comparison to conventional amphotericin B or itraconazole was evaluated with a decision tree model followed by a life-time Markov model, focusing on long-term survival of patients treated for invasive aspergillosis. Efficacy after 12 weeks of treatment from clinical trials was used to estimate long-term effectiveness by extrapolating these short-term results over time. Information on medical resource consumption, treatment pathways and switch proportions were obtained from both the literature and Experts. Probabilistic analysis was used to compare the cost-effectiveness among the regimens. RESULTS: With voriconazole, the mean cost for treating invasive aspergillosis per patient was E32 651 (2.5th percentile and 97.5th of uncertainty distribution: E30 037; E36 859), compared to E33 616 (E30 920; E39 633) for conventional amphotericin B and E29 115 (E23 537; E61 414) for itraconazole. The mean survival of patients treated with voriconazole was 174.0 life weeks (160.1; 188.8), compared to 116.1 life weeks (104.8; 128.0) for conventional amphotericin B and 150.4 life weeks (109.1; 194.4) for itraconazole. The beneficial effects of voriconazole on both cost and effectiveness compared with conventional amphotericin B resulted in a probability of 69.8% that voriconazole was a dominant treatment (i.e. less costs and longer survival). The incremental cost-effectiveness ratio of voriconazole versus itraconazole was E150 per life week (i.e. 7800 euros per life-year gained). Depending on the willingness to pay (WTP) the probability of being cost-effective vs. itraconazole increased to a maximum probability of 70%. CONCLUSION: In the treatment of invasive aspergillosis, voriconazole is dominant over amphotericin B and cost-effective in comparison to itraconazole.

Aphid parasitoid complex in potato in the context of IPM in Belgium.

Parasitic hymenoptera of potato aphids were collected and identified in 2000 and 2001 in 11 potato fields in Belgium. Nine Aphidiidae species, parasiting seven aphid species, were found: Aphidius ervi, Aphidius matricariae, Aphidius picipes, Binodoxys angelicae, Diaeretiella rapae, Praon abjectum, Praon gallicium, Praon volucre and Toxares deltiger. A. ervi and A. picipes were the dominant and sub-dominant species, with 54% and 28% of the primary parasite collected, respectively. Both species and T. deltiger were found on Aphis nasturti, Aulacorthum solani, Macrosiphum euphorbiae and Myzus persicae, the four most important aphid potato in Belgium. Parasitism rate of A. nasturtii and, to a lesser extent, M. euphorbiae was low compared to A. solani and M. persicae. Parasitism of A. solani was particularly high, with 63.5% in 2000 and 89.2% in 2001, and this species was the preferred host of several Aphidiid species. The abundance of alternative hosts as other crops pest aphids or non-pest aphids on wild plants in agroecosystems could explains the efficacy of these species. The biological control of A. nasturtii and M. euphorbiae by parasitic hymenoptera was poorer, and several studies need to be undertaken to find suitable parasitic hymenoptera species effective on these aphids and agro-environmental measures able to promote them.

Selectivity lists of pesticides to beneficial arthropods for IPM programs in carrot--first results.

In order to improve IPM programs in carrot, 7 fungicides, 12 herbicides and 9 insecticides commonly used in Belgium were tested for their toxicity towards five beneficial arthropods representative of most important natural enemies encountered in carrot: parasitic wasps - Aphidius rhopalosiphi (De Stefani-Perez) (Hym., Aphidiidae), ladybirds - Adalia bipunctata (L.) (Col., Coccinellidae), hoverfly - Episyrphus balteatus (Dipt.. Syrphidae), rove beetle - Aleochara bilineata (Col., Staphylinidae) and carabid beetle - Bembidion lampros (Col., Carabidae). Initialy, all plant protection products were tested on inert substrate glass plates or sand according to the insect. Products with a corrected mortality (CM) or a parasitism reduction (PR) lower than 30% were kept for the constitution of positive list (green list). The other compounds were further tested on plant for A. rhopalosiphi, A. bipunctata, E. balteatus and soil for B. lampros and A. bilineata. With these extended laboratory tests results, products were listed in toxicity class: green category [CM or PR < or = 30%], yellow category [30% < CM or PR < or = 60%] and orange category [60% < CM or PR < or = 80%]. Products with toxicity higher than 80% on plants or that reduce parasitism more than 80% on soil were put in red category and are not recommended to Integrated Pest Management programs in carrot. Results showed that all fungicides tested were harmless to beneficials except Tebuconazole, which was slightly harmful for A. bipunctata. Herbicides were also harmless for soil beneficials, except Chlorpropham. This product was very toxic on sand towards A. bilineata and must be tested on soil. All soil insecticides tested were very toxic for ground beneficials and considered as non-selective. Their use in IPM is subject to questioning in view of negative impacts on beneficials. Among foliar insecticides, Dimethoate and Deltamethrin are not recommended for IPM because their high toxicity for all beneficials. The other foliar insecticides were more selective; any of them were harmless for all species tested.

Dose-response relations between occupational exposures to physical and psychosocial factors and the risk of low back pain.

AIMS: To assess dose-response relations between occupational exposures to physical and psychosocial factors and the risk of low back pain. METHODS: A cohort of 523 subjects, working in nursing homes and homes for the elderly, was followed prospectively for one year. Physical load for different occupations was assessed by quantitative observations at the workplace. Information on low back pain and other factors was gathered with questionnaires administered at baseline and at one year. Two outcome measures of low back pain incidence were used: any new episode of pain lasting for at least a few hours during follow up (LBP); and any new episode of disabling pain that interfered with daily activities during follow up (LBP/D). Hierarchical regression analysis with a spline function was used to estimate dose-response relations. RESULTS: The risk of LBP was not associated with physical factors, controlling for confounders; but this outcome was inversely associated with age and weakly, though imprecisely, associated with two psychosocial factors--low decision authority and high work demands. In contrast, the risk of LBP/D was positively associated with age and not associated with the psychosocial factors. Trunk flexion over 45 degrees was monotonically associated with the risk of LBP/D; the estimated relative risk was 3.18 (95% CI 1.13 to 9.00) for 1 hour and 45 minutes of bending per week (90th centile), relative to 30 minutes per week. The hierarchical estimates of effect were more stable than were the maximum likelihood estimates. CONCLUSION: Occupational exposure to trunk flexion over 45 degrees appears to be a risk factor for low back pain with disability among persons employed in nursing homes and homes for the elderly in the Netherlands.

Aphid specific predators in potato in Belgium.

In summer 2003, aphid specific predators populations were sampled in four potato fields located in Wallonia. Ladybirds, hoverflies and lacewings were the most abundant predators collected. A total of 6 ladybird, 8 syrphid and 2 lacewing species was found. Predator species composition and abundance greatly differed from field to field, while aphids populations were more or less similar with a peak of 3 to 5 aphid/potato leaf in mid-July. All chrysopidae sampled belong to the Chrysoperla kolthoffi (NAVAS) species except 2 specimens of Chrysopa perla L. Episyrphus balteatus DEGEER was the dominant syrphid species. Sphaerophoria scripta L. and Syrphus vitripiennis MEIGEN were subdominant species and 1 or 2 specimens of Melanostoma scalare F., Melanostoma mellinum (L.), Melliscaeva cinctella (ZETTERSTED), Metasyrphus corollae (F.) and Platycheirus albimanus F. were also collected. Coccinella septempunctata (L.) and Propylea quatuordecimpunctata (L.) were the dominant ladybird species. P. 14-punctata was proportionally more abundant in beating samples than with visual inspection samples, indicating that this ladybird is probably more discrete than C. septempunctata and that their populations can be underestimated with only visual inspection. Adalia bipunctata (L.) was also present in nearly all samples, but with lower levels of populations. Coccinella quinquepunctata (L.) and Adalia tenpunctata (L.) were found in small numbers at one occasion. Larvae and adults of Harmonia axyridis PALLAS, the multicoloured Asian beetle introduced in glasshouse for aphid control, were found in 3 fields out of 4, with 1.5 to 6% of ladybirds sampled. This is the first record of H. axyridis in open field in Belgium. As this beetle has been found in numbers at different locations and with different growth stages including larvae and pupae, it is supposed that H. axyridis is adapted to ware potato aphid and able to complete a full life cycle in this crop. As this species is very competitive for exploitation of food resources and aggressive for other predators, they are several questions about the evolution in the future of aphid specific predator biodiversity in ware potato.

Drawing-up of pesticide selectivity lists to beneficial arthropods for IPM programmes in potato.

In order to promote IPM programmes in potato, the toxicity of 19 fungicides, 4 herbicides and 11 insecticides commonly used in this crop in Belgium was assessed on three beneficial arthropods. These species were representative of the most important aphid specific natural enemies encountered in potatoes: a parasitic wasp--Aphidius rhopalosiphi (De Stefani-Perez) (Hym., Aphidiidae), a ladybird--Adalia bipunctata (L.) (Col., Coccinellidae) and a hoverfly--Episyrphus balteatus (Dipt., Syrphidae). In a first time, pesticides were tested on glass plates on A. rhopalosiphi adults and A. bipunctata and E. balteatus larvae. For each insect, products inducing corrected mortality (Mc) lower than 30% were directly classified in a positive list for harmless products (green list). The other compounds were further tested on plants and listed in toxicity classes according to mortalities induced during this extended laboratory test: harmless (Mc < 30%), slightly harmful (30% < Mc < 60%), moderately harmful (60% < Mc < 80%) and harmful (Mc > 80). A chemical determination of pesticides residues was also performed for each experiment in order to determine the exposure of beneficial arthropods to pesticide residues and to validate the application of chemicals on tested substrates. On the basis of the results of acute toxicity tests, the period of each pesticide use according to normal agricultural practices and the abundance and importance of the three different groups of aphid natural enemies at different periods of the year, four pesticides lists were built up. Each list corresponded to a different period of pesticides application: Period I--from seedling to beginning of June (based on A. rhopalosiphi tests), Period II--beginning to end of June (based on A. rhopalosiphi tests), Period III beginning to end of July (based on E. balteatus and A. bipunctata tests) and Period IV--August to harvest (no exposure of beneficials). Results showed that herbicides were not toxic to the three species and can be used according to normal agricultural practices without restrictions. All fungicides can also be used without restrictions at recommended rates. Only the mixture Metalaxyl-M + Fluazinam was slightly harmful to A. bipunctata but had no effects on A. rhopalosiphi and E. balteatus. Results were more contrasted for insecticides and none of them was totally selective for all the 3 beneficial arthropods. Therefore, they can only be used with restrictions at periods II and III, according to the beneficial species that need to be protected.