Datasets:

doriab
/

CareMedEval

text stringlengths 0 11k
Maternal Consumption of Coffee and Caffeine-containing Beverages and Oral Clefts: A Population-based Case-Control Study in Norway
Anne Marte W Johansen, Allen J Wilcox, Rolv T Lie, Lene F Andersen, Christian A Drevon

Abstract
A large, population-based case-control study of facial clefts was carried out in Norway between 1996 and 2001. The study included 573 cases—377 with cleft lip with or without cleft palate and 196 with cleft palate only—and 763 randomly selected controls. Maternal consumption of coffee and other caffeine-containing beverages in early pregnancy was recorded shortly after birth. Compared with that for no coffee consumption, the adjusted odds ratios for cleft lip with or without cleft palate were 1.39 (95% confidence interval: 1.01, 1.92) for less than 3 cups a day and 1.59 (95% confidence interval: 1.05, 2.39) for 3 cups or more. Coffee consumption was not associated with risk of cleft palate only (for ≥3 cups vs. none, adjusted odds ratio = 0.96, 95% confidence interval: 0.55, 1.67). Tea consumption was associated with a reduced odds ratio of both cleft lip with or without cleft palate and cleft palate only. There was little evidence of an association between caffeine exposure and clefts when all sources of caffeine were considered. Adjustment for known confounding factors in general had minor effects on risk estimates. Still, the authors could not rule out the possibility of uncontrolled confounding by factors associated with the habit of drinking coffee.

Keywords: caffeine, cleft lip, cleft palate, coffee, pregnancy


Orofacial clefts are among the most common birth defects, and the prevalence in Norway (2.2 per 1,000 livebirths) is particularly high. The etiology of clefts is complex and largely unknown. The high risk of recurrence of clefts among first-degree relatives (as much as 56 times the background prevalence) suggests a strong genetic component (1). However, environmental factors such as maternal nutrition (2–4), vitamin supplements (5), smoking (6), and binge alcohol consumption (7) also appear to contribute.

Coffee consumption is relatively high in Norway. In the latest nationwide study, average coffee intake was half a liter a day, with peak intake among those aged 40–60 years (8). Data from animal studies suggest that large single doses of caffeine may cause palatal clefts as well as other birth defects (9), although studies of coffee consumption in humans have provided little evidence of teratogenic effects (10). In a systematic review of 3 studies of orofacial clefts and caffeine, high coffee intake was associated with a slight increase in risk (11). According to a recently published article, maternal caffeine intake during pregnancy was associated with fetal growth restriction (12).

Coffee is a commonly consumed beverage among pregnant women, and even a small increase in malformation risk could be a matter of concern. We used data from a population-based case-control study to evaluate the association of maternal consumption of coffee and caffeinated beverages in early pregnancy with the risk of delivering an infant with an orofacial cleft.


MATERIALS AND METHODS
All infants born with facial clefts in Norway are treated at government expense in surgical centers at university hospitals located in Oslo and Bergen. In collaboration with these 2 centers, we identified all babies born from 1996 to 2001 who were referred for treatment for either cleft lip with or without cleft palate (CLP) or cleft palate only (CPO). Controls were recruited during the same period by randomly selecting approximately 4 births per 1,000 from the National Medical Birth Registry (which includes all births in the country). These births served as controls for both case groups, with the target of 2 controls per case of CLP (nearly 4 controls for each case of CPO). Both cases and controls were recruited during their first weeks of life.

Our present study was approved by the regional ethics review board, the Norwegian Data Inspectorate, and the US National Institute of Environmental Health Sciences Review Board. All participating mothers provided informed consent.

Participants
There were 676 women in Norway who delivered infants requiring surgery for orofacial clefts during 1996–2001. We excluded 24 mothers who did not speak Norwegian or whose infant died after birth, leaving 652 eligible mothers. Of these, 88% (n = 573) agreed to participate. We randomly selected 1,022 control mothers of livebirths via the national Medical Birth Registry within 6 weeks of delivery. After excluding 16 who were not Norwegian speakers or whose infant died, 1,006 mothers of controls were eligible, of whom 76% (n = 763) agreed to participate. Data on intake of coffee and other caffeinated beverages were available for all participating mothers of cases and controls. We identified other birth defects among the cleft cases by using 3 data sources: the Medical Birth Registry (based on delivery records and hospital records from the first week of life), medical records at the hospital performing the corrective surgery, and mother's questionnaire. Accompanying defects were reported for 17% of CLP cases and 40% of CPO cases. Cases of clefts without accompanying defects have been categorized as isolated clefts.

Data collection
Mothers completed a 32-page questionnaire covering demographic characteristics, reproductive history, and exposures during pregnancy (including smoking, alcohol consumption, coffee intake, medication use, and occupational and household exposures). Median time from delivery to completion of the questionnaire was 14 weeks for cases and 15 weeks for controls. The questionnaire included items on maternal consumption of caffeine-containing beverages (coffee, tea, and soft drinks) during the first 3 months of pregnancy. For each beverage, there was 1 question with 5 response categories: none, number of cups per day, number of cups per week, number of cups per month, and number of cups per year (without specifying the size of the cup). An English translation of the questionnaire is available online (http://www.niehs.nih.gov/research/atniehs/labs/epi/studies/ncl/ncl_pregnancy_en.pdf).

Statistical methods
The risk of delivering offspring with an orofacial cleft was estimated by odds ratios with 95% confidence intervals in unconditional logistic regression models. All beverages were summarized and were analyzed in the same way, as follows. “Cups per day” was computed from reported number of cups consumed per day, per week (divided by 7), or per month (divided by 30). Women who reported consuming less than 1 cup per month were categorized as consuming zero. The cups-per-day variable was analyzed as a continuous variable. Finally, a 3-category variable was created: 0 cups per day (reference category), more than 0 but less than 3 cups per day, and 3 or more cups per day. Trends across the categories were evaluated, with zero as the reference.

An estimate of caffeine from all sources was computed from the data on coffee, tea, and caffeinated soft drinks. Caffeine content was estimated as 100 mg per cup of coffee, 40 mg per cup of tea, and 20 mg per cup of caffeinated soft drink based on values from the Norwegian Health Authorities Web page (http://www.matportalen.no/Emner/Gravide (in Norwegian)). Risk of clefts was evaluated per 100-mg increase in caffeine intake (continuous variable) and for the categories >100–<500 mg and ≥500 mg of caffeine relative to 0–100 mg.

Adjustments were made for potential confounders (factors associated with clefts in other studies, most of which were also associated in our study), namely, dietary vitamin A (quartiles), dietary folate (quartiles), folic acid supplement (400 μg/day, yes or no), vitamin supplement use (yes or no), consumption of alcohol in early pregnancy (number of drinks per sitting), smoking (ordinal linear with 5 categories: none; passive only; and 1–5, 6–10, and ≥11 cigarettes a day), nausea during the first trimester (yes or no), employment in early pregnancy (yes or no), education (ordinal linear with 6 categories), father's income (ordinal linear with 3 categories), and year of birth. Evaluations of possible interactions with coffee intake were carried out for use of folic acid supplements and smoking. In evaluating the effects of coffee or tea separately, we adjusted for the other (categorized as number of cups per day). Because CLP and CPO are considered etiologically distinct outcomes, we conducted separate analyses for each (1). Separate analyses were also performed for isolated clefts (i.e., excluding those with accompanying defects). Because oral clefts are relatively rare birth defects, odds ratios are close approximations of relative risks. All statistical analyses were performed by using SPSS 14.0 software (SPSS Inc., Chicago, Illinois).

RESULTS
Women who gave birth to infants with CLP were taller, less educated, and less likely to work during the first trimester compared with mothers who gave birth to healthy controls (Table 1). Compared with mothers of controls, fewer mothers of CLP cases used a folic acid supplement, and they were more often coffee consumers and smokers. Fewer mothers of CLP and CPO cases drank tea compared with mothers of controls (Table 1).

Maternal coffee consumption was associated with an increased risk of CLP. In the adjusted analyses, the odds ratio of CLP increased by 7% per-cup increase in daily coffee intake (adjusted odds ratio = 1.07, 95% confidence interval (CI): 1.00, 1.16). Compared with those for women with zero coffee consumption, the adjusted odds ratios of CLP were 1.39 (95% CI: 1.01, 1.92) for daily coffee consumption of less than 3 cups a day and 1.59 (95% CI: 1.05, 2.39) for consumption of 3 or more cups a day (Table 2), and inspection of categories of coffee intake confirmed that there was a trend in risk by dose (Ptrend = 0.013 in the adjusted analyses). The association between coffee consumption and CLP persisted among nonsmokers (among whom there would presumably be no residual confounding by smoking) and among mothers of isolated cleft cases (Table 2). We found no evidence of an association between maternal coffee consumption during the first trimester and the risk of CPO (Table 2).

Consumption of caffeine-containing tea was associated with a decrease in the odds ratio of both CLP and CPO (Table 2). Compared with no tea intake, daily tea intake of 3 or more cups gave adjusted odds ratios of 0.55 (95% CI: 0.32, 0.95) for CLP and 0.58 (95% CI: 0.31, 1.07) for CPO. Soft drinks that contain caffeine were positively associated with both types of facial clefts, although with confidence intervals that did not exclude 1 (Table 2).

Table 3 shows maternal intake of caffeine from all beverages in relation to cleft risk. Although there was a positive association in the crude analysis, the association was reduced after adjustment. We also considered mothers who reported drinking coffee during the year before pregnancy but not during the first trimester. There was only a weak, positive association with risk of CLP in this group (adjusted odds ratio = 1.21, 95% CI: 0.80, 1.85) (Table 4).


DISCUSSION
Maternal intake of coffee during the first 3 months of the pregnancy was associated in a dose-response manner with risk of delivering an infant with CLP. In contrast, we found no evidence of an association between coffee intake and risk of CPO. The association between coffee and CLP was only slightly decreased by adjustments for confounders, and it persisted for nonsmokers and the subset of infants with isolated CLP.

There is no known mechanism by which coffee intake might increase the risk of CLP. Effects on homocysteine is one potential pathway. Evidence suggests that maternal hyperhomocysteinemia may be linked to increased risk of CLP (13). Coffee intake increases the plasma concentration of homocysteine (14–16), as does smoking (15), a well-established risk factor for CLP (17). Folic acid supplements are consistently associated with reduced risk of CLP (18), and such supplements reduce plasma homocysteine (15). In the present study, the association between high coffee intake and CLP was slightly stronger for women who did not take a folic acid supplement (crude odds ratio = 1.99, 95% CI: 1.30, 3.04), although this interaction did not approach statistical significance (P for interaction = 0.89).

Consumption of caffeine-containing tea was strongly associated with a reduced risk of both CLP and CPO. Potential health benefits of tea have been linked to its high content of antioxidants (19). This is not a likely explanation for the reduced cleft risk, however, because coffee has an even higher total content of antioxidants than tea does (20). The association of tea with lower risk of both types of clefts may reflect the presence of unmeasured confounding factors. Tea drinkers may have more healthy habits; consumers of high quantities of tea in our study were older, more educated, and drank less alcohol per sitting compared with nonconsumers. Furthermore, the diet of consumers of high quantities of tea contained more of every macronutrient and almost every micronutrient compared with the diet of nonconsumers (data not shown).

The association of coffee with CLP appears to be independent of the role of caffeine. We found no association between total caffeine intake and risk of CLP (reflecting the combination of the positive association with coffee and the negative association with tea). Only 5 women drank decaffeinated coffee exclusively, too few for a separate analysis. When we combined data on decaffeinated coffee with those on caffeinated coffee, the odds ratios for CLP were slightly increased, consistent with the association of coffee and CLP being unrelated to caffeine intake (data not shown).

By the same token, high-quantity coffee consumers (>3 cups/day) were older, less educated, smoked more, and drank more alcohol per sitting compared with nonconsumers. Even though we adjusted for possible confounding factors, we cannot rule out the possibility of residual confounding by known factors such as smoking or confounding by unmeasured factors. Although we did not control for diet, there was no evidence of a poorer diet among consumers of high quantities of coffee, who actually reported eating more vegetables, less sugar, and more of many essential nutrients (data not shown).

Another potential source of bias in this study comes from possible selective participation of controls. While participation was relatively high (88% for cases and 76% for controls), the lower participation of controls leaves room for differential participation by coffee drinking. However, it is unlikely that such selection would occur for one type of facial cleft and not the other. Coffee was strongly associated with CLP but not with CPO in our data. This specificity has been found for other factors associated with facial clefts (6). By the same argument, the lack of specificity for the tea association (which was protective for both types of facial clefts) raises the question of whether this association might be due to some unmeasured bias.

Women may reduce their coffee consumption as they develop pregnancy symptoms of nausea and vomiting. Given that closure of the lip and palate occurs relatively late in the first trimester (8–12 weeks after the last menstrual period) (1), it is possible that reported coffee intake is greater than actual consumption at the crucial stage of embryonic development. This bias would tend to weaken the observed association.

Our study had the advantage of enrolling mothers soon (an average of 15 weeks) after delivery. In addition, questions on coffee and other caffeinated beverages constituted only a small part of the questionnaire, with no specific emphasis on these beverages that would be expected to bias reporting. Another strength of the study was collection of extensive data on potential confounders such as folic acid supplement use, alcohol consumption, and smoking. Confounders were adjusted for as continuous variables in our analyses, but adjustment using categorized variables did not alter the results.

A meta-analysis of 3 studies on maternal coffee consumption and orofacial clefts found a slight increase in the risk of clefts (11). All 3 studies used 0 cups as the reference category. When high coffee intake was compared with low intake, the pooled odds ratio was 1.2 (95% CI: 0.9, 1.6). One of the studies provided results for CLP separately, with an adjusted odds ratio for CLP of 1.3 (95% CI: 0.9, 1.9) associated with more than 3 cups of coffee a day (11). The 2 other studies combined both types of clefts in their analyses. In the report by Kurppa et al. (21), drinking more than 4 cups of coffee a day had no association with risk of clefts (unadjusted odds ratio = 1.0, 95% CI: 0.6, 1.6), whereas, according to McDonald et al. (22), more than 3 cups of coffee daily resulted in an adjusted odds ratio of 1.4 (95% CI: 0.7, 2.7).

A recent cohort study from Denmark found that coffee intake was associated with a reduced risk of CLP (23). The authors observed an odds ratio of 0.66 (95% CI: 0.27, 1.62) when comparing daily intake of more than 5 cups of coffee with no intake. The Danish study had the advantage of a prospective design, although one cost of this design was a relatively small number of cases (134 with CLP).

In summary, results from our study showed a dose-dependent association between coffee consumption during the first trimester and increased risk of CLP. This association was specific to CLP, with no association found between coffee consumption and risk of CPO. There was little or no evidence for an association between caffeine from other types of beverages and CLP. Even with extensive adjustments for confounders, we cannot eliminate the possibility that the associations between coffee consumption and risk of CLP are due to uncontrolled confounding by factors associated with the habit of drinking coffee. Still, considering our results and the prior mixed evidence for a coffee effect on clefts, women cannot be assured that maternal coffee consumption is entirely benign for the developing fetus. Other clefts studies now in progress should give close attention to a possible association of coffee consumption with facial clefts.
Assessment of Glenohumeral
Subluxation in Poststroke Hemiplegia:
Comparison Between Ultrasound and
Fingerbreadth Palpation Methods

Praveen Kumar, Marianne Mardon, Michael Bradley, Selena Cray,
Annette Swinkels

Background. Glenohumeral subluxation (GHS) is a common poststroke compli
cation. Treatment of GHS is hampered by the lack of objective, real-time clinical
measurements.
Objective. The aims of this study were: (1) to compare an ultrasound method of
GHS measurement with the fingerbreadth palpation method using a receiver oper
ating characteristic curve (ROC) and (2) to report the sensitivity and specificity of this
method.
Design. A prospective study was conducted.
S etting. The study was conducted in local hospitals and day centers in the
southwest of England.
Patients. One hundred five patients who had one-sided weakness following a
first-time stroke (51 men, 54 women; mean age=71 years, SD=11) and who gave
informed consent were enrolled in the study.
Measurements. Ultrasound measurements of acromion-greater tuberosity
(AGT) distance were used for the assessment of GHS. Measurements were under
taken on both shoulders by a research physical therapist trained in shoulder ultra
sound with the patient seated in a standardized position. Fingerbreadth palpation
assessment of GHS was undertaken by a clinical physical therapist based at the
hospital, who also visited the day centers.
Results. The area under the ROC curve was 0.73 (95% confidence interval [95%
Cl] =0.63, 0.83), suggesting that the ultrasound method has good agreement com
pared with the fingerbreadth palpation method. A cutoff point of £0.2 cm AGT
measurement difference between affected and unaffected shoulders generated a
sensitivity of 68% (95% CI=51%, 75%), a specificity of 62% (95% CI=47%, 80%), a
positive likelihood ratio of 1.79 (95% CI=1.1, 2.9), and a negative likelihood ratio of
0.55 (95% CI=0.4, 0.8).

End of preview. Expand in Data Studio

CareMedEval dataset

CareMedEval dataset: Evaluating Critical Appraisal and Reasoning in the Medical Field

	Links
Repository	GitHub, HuggingFace
Paper	https://arxiv.org/abs/2511.03441
Contact	Doria BONZI, Alexandre GUIGGI

Description

CareMedEval (Critical appraisal and Reasoning Medical Evaluation) is a French Multiple Choice Question Answering (MCQA) dataset focused on evaluating critical appraisal skills in the medical field for scientific articles, as practiced in French medical education.

This dataset is composed of 534 questions taken from Lecture Critique d'Articles (LCA, "Critical Appraisal of research Articles) exams, taken by sixth-year medical students in France. During the LCA exams, students are asked to answer a series of multiple-choice questions grounded in the critical reading of a given scientific article.

Source data

This dataset is built from two main sources:

Epreuves Classantes Nationales (ECN) website, where we can find the official national LCA exams
Collège National des Enseignants de Thérapeutique (CNET) website, which publishes mock LCA exams reviewed and approved by the CNET's educational committee to ensure alignment with real LCA exams.

The articles used in these exams are publicly available online.

Figure. CareMedEval dataset constitution pipeline.

Languages

The scientific articles are in English. The questions, answers, and justifications are originally in French. You can find in ./english_only/ a version of this dataset with automatically translated questions, answers and justifications, from French to English. We used Gemini 2.5 Flash for this translation.

Annotations and labels

Each question in the CareMedEval dataset was manually annotated with one or more labels, reflecting the cognitive and analytical skills required to answer it. Question labeling was conducted by one annotator with a background in general medecine, using reference textbooks and their medical expertise. The labels are defined as follows:

Label	Description	Skills required
design	Identification of study design	Information retrieval
statistics	Understanding and interpretation of statistics	General knowledge, Information retrieval
methodology	Knowledge of scientific methodology	General conceptual understanding
limitations	Critical appraisal of biases and limitations	Contextual reasoning
applicability	Clinical relevance and applicability	Contextual reasoning

Since questions often target multiple dimensions of skills, they can be assigned more than one label. These labels can help to determine if certain categories of questions are more challenging than others for models, depending on the types of skills they require.

For data sourced from ECN website, we manually corrected the exam questions with the help of a general practitioner, as the official answer keys were not publicly available for these past exams.

For data from CNET website, we also collected the correction and justifications provided by medical professionnals for some questions. These justifications explain why some answers are correct or false, offering valuable insights into clinical reasoning and critical appraisal.

Dataset structure

To access the full dataset, go to ./data/all/, which contains both /articles/ and /questions.json.
To access only the questions that include human justifications (204 questions out of 534), go to ./data/with_justifications/.
To access our manually annotated subset of 16 questions indicating whether questions require context (articles) to be answered correctly, go to ./data/requires_context/.

Questions

Example of a question:

        "id": "fd30e0bb8cbe1d73039ffebab8e98019836a2e1af03b383f48be020d79f61310",
        "id_article": "article_3",
        "source_exam": "https://therap.fr/lca-mai-2019/",
        "date_exam": "2019",
        "article_link": "https://jamanetwork.com/journals/jama/fullarticle/2547755",
        "article_date": "2016",
        "question": "La publication de cet essai négatif :",
        "answers": {
            "a": "Est une situation rare les essais négatifs sont le plus souvent non publiés",
            "b": "Est inutile puisque n’apporte aucune information sur le traitement en pratique",
            "c": "Doit faire cesser toute investigation ultérieure sur cette molécule dans cette indication",
            "d": "Permettra d’intégrer plus facilement cet essai dans une revue systématique/metaanalyse",
            "e": "Est logique la publication des résultats de tous les essais est obligatoire"
        },
        "correct_answers": [
            "a",
            "d"
        ],
        "essential_answers": [],
        "unacceptable_answers": [],
        "labels": [
            "limitations",
            "applicability"
        ],
        "justification": "50% des essais sont non publiés et ce surtout s'ils sont négatifs. Savoir qu’un traitement n’est pas supérieur à un autre est en soi une information importante. Voir la discussion de l’essai. Une revue systématique pour donner un reflet réel de l’ensemble de la recherche sur une question doit inclure tous les essais réalisés qu’ils soient positifs ou négatifs, sinon l’éventuelle méta-analyse donnera un résultat biaisé. Il est plus simple de trouver les résultats si l’essai est publié. Hautement souhaitable et recommandée.",
        "nb_correct_answers": 2

Each question in the dataset is represented as a JSON object with the following fields:

Field	Description
`id`	A unique identifier for the question
`id_article`	An internal ID linking the question to the corresponding article in the dataset
`source_exam`	The URL of the exam or online resource (e.g., CNET or ECN) from which the question was extracted
`date_exam`	The date on which the exam was presented
`article_link`	The direct link to the full-text scientific article used as the basis for the question
`article_date`	The date on which the article was published
`question`	The multiple-choice question (MCQ) as presented in the exam
`answers`	A dictionary mapping each option label (A to E) to its full answer text
`correct_answers`	A list of option labels corresponding to the correct answers. Questions can have one or multiple correct options
`essential_answers`	A list of essential answers. If any of these are not selected, the question is considered incorrect in LCA grading
`unacceptable_answers`	A list of inadmissible answers. If any of these are selected, the question is considered incorrect due to the severity of the error in LCA grading
`labels`	A list of strings, indicating the types of reasoning or knowledge required to answer the question (see label table above)
`justification`	An expert-written explanation justifying the correct (and sometimes incorrect) answers; available for CNET-sourced questions
`requires_context`	Boolean indicating if the question requires the article to be answered correctly or not
`nb_correct_answers`	The number of correct options for the question

Articles

Each question in the dataset is linked to a scientific article through the id_article field, which serves as a reference key to the corresponding article file. Articles are available in plain text (.txt) format for easier processing, and the original PDF versions are also included. In addition, each question includes a direct article_link field pointing to the online HTML version. To provide a more concise context, we extracted the abstracts of the articles and stored them in ./data/all/articles_abstract, available in plain text (.txt) format.

Dataset benchmark

Figure. Overview of the model evaluation pipeline of the CareMedEval dataset.

We conducted a series of experiments on our dataset to evaluate the critical appraisal skills of a diverse pool of LLMs. While our dataset can support multiple experimental setups, we focus here on a MCQA task for reference results, based solely on the textual content of the articles. Details on our experimental protocol are available on TBA.

Results

Citation

To cite this dataset:

@misc{bonzi2025caremedevaldatasetevaluatingcritical,
      title={CareMedEval dataset: Evaluating Critical Appraisal and Reasoning in the Biomedical Field}, 
      author={Doria Bonzi and Alexandre Guiggi and Frédéric Béchet and Carlos Ramisch and Benoit Favre},
      year={2025},
      eprint={2511.03441},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2511.03441}, 
}

Downloads last month: 152

Size of downloaded dataset files:

2.23 MB

Size of the auto-converted Parquet files:

1.32 MB

Number of rows:

9,270