During pregnancy, there are some common ailments many women experience. Over-the-counter medications (OTC-medications) are widely used to treat these ailments. Because of their wide use, some OTC-medications (e.g. acetaminophen) are considered to be safe for pregnant women. However, by the absence of randomized clinical trials, the knowledge about the safety of OTC-medication exposure during pregnancy is dependent on observational studies. The only OTC-medications that are well-known to be potentially harmful are Non-Steroidal Anti-Inflammatory Drugs (NSAIDs). Fetal NSAID exposure is associated with an increased risk of constriction of the ductus arteriosus (DA) [1] and the risk of spontaneous abortion [2]. It is particularly difficult to investigate the consequences of OTC-medication use on fetal outcomes, because their use is not fully documented. Even healthcare providers are not always aware of OTC-medication use by their patient during pregnancy. In the Dutch self-care guidelines for healthcare professionals [3] most OTC-medications are discouraged to use during pregnancy because of the lack of evidence for either safety or fetal risk. Therefore, healthcare providers should provide personalized counseling to pregnant women about possible negative side effects of OTC-medication on their unborn child. In order to achieve this it is preferable to indicate women that are at risk for OTC-medication use.

The study results published so far on OTC-medication use during pregnancy are difficult to compare, because research populations and study designs differ strongly. For example, studies from the United States found percentages of women that self-medicated with OTC medication varying between 92,6% [4] and 53,6% [5]. Furthermore, drug utilization patterns vary strongly between different countries. It is therefore important to study the prevalence and predictors of OTC-medication use in a Dutch setting. Previous studies already investigated drug use during pregnancy in the Netherlands, but they used different study designs and did not focus on OTC-medication [6, 7]. By describing possible predictors of OTC-medication use during pregnancy, midwifes and obstetricians in the Netherlands can be assisted in identifying pregnant women that are likely to use OTC-medication. In this population ‘at-risk’, personalized counseling can be given in order to mitigate medication-related risks for the unborn child. We have implemented the predictors in a prediction model applicable in daily practice.

We found that the prevalence of OTC-medication use in a Dutch cohort of pregnant women was 12.5%. Five predictors including nulliparity, prescription drug use, comorbidity, BMI and GP visits could be identified and these predictors were used to design a prediction model which is easy to implement in clinical practice. The prediction model assists healthcare professionals in identifying pregnant women that are likely to use OTC-medication, so information can be given directly to the women at risk.

Although it is difficult to compare because of the divergence in study designs and study populations, the prevalence of 12.5% is low in comparison to similar studies that have been published [4, 5, 12, 13]. An Irish study found a prevalence of 19.5% [13] (2010) and studies from the United States found percentages of women that self-medicated with OTC-medication varying between 92,6% [4] (2003) and 53,6% [5] (1993). A recent study among Hispanic women from the United States found a prevalence of 23% [15] (2010). A Dutch study which was published in 1991 found a prevalence of 45% [16]. Because these data were collected 21 years ago, the population is not representative for the population nowadays. It is likely that pregnant women today are more aware of the risks of medication use during pregnancy and this could explain the low prevalence of OTC-medication use in our study. Another explanation for the large difference in prevalence between our study and the study from 1991 is that de Jong-van den Berg et al. assembled data by an interview of approximately 30 minutes and medication use was ascertained in detail in three different ways. Besides that, all vitamins -except folic acid- were included in their study whereas we excluded vitamin D users. Other studies on drug use during pregnancy performed in the Dutch population are not comparable to our study, because they used data collected from the pharmacy information system or made no distinction between prescribed and self-administered drugs [6, 7].

Previous studies described that the use of OTC-medication was positively related to multiple illnesses [4, 5], Caucasians [4, 5, 12], women with more than a high school education [12], women who were at least 20 years of age [12, 13], nulliparity [13], smoking [13, 14] and being single [13]. We included all these variables in our study, except for marital status, because those data were not available. Besides the five predictors previously mentioned, we also found a non-significant higher frequency of OTC-medication use in non-Western women compared to Western women. Non-Western women were underrepresented (n = 26) in the study due to the exclusion criteria. Therefore, the predictor was not strong enough to include in the multivariate model. We encountered the same power problem for the association between OTC-medication use and the variables alcohol use (n=6), previous prematurity/child mortality (n = 7) and low education (n=7). Therefore, the study results should be interpreted with caution for women with (one of) these characteristics.

Although our results partially differ from the predictors published so far, we were able to design a strong prediction model, with five predictors that independently contribute. Further research with larger cohorts is necessary to study the non-significant associations. The score (Table 5) and the equation (Equation 1) are designed for use in daily practice. Women with a score of 8 or higher have 30.1%-57.4% chance of using OTC-medication during pregnancy. 5.4% of the women included fell into this category. 63% of the women had a score between 2 and 7 and 31.6% of the women had a score between 0 and 1. Explorative analysis showed that the rule performed more or less similar results for subgroups of medications. We suggest the women with a score of 8 or higher, are eligible for comprehensive information provision. The sensitivity and specificity of the model at this cut-off score are 0.122 and 0.956, respectively. This means that the chance of a false positive result is small, while the chance of a false negative result is large. Sensitivities and specificities of the model at other cut-off values are represented in Table 5. The choice of a cut-off score is arbitrary and healthcare providers should make their own considerations concerning the information provision to pregnant women that are at risk for OTC-medication use.

Our study is the first in identifying predictors of OTC-medication use in the Dutch population. Other strong points of our study are the large cohort size (n=1246) and the completeness of the data concerning a variety of socio-demographic, obstetric and medical characteristics. Because our dataset was primarily assembled for a study with another objective and because the questionnaire was filled out prior to delivery, the reporting of OTC-medication use was relatively unbiased. Another strong point of our study is that data were collected by self-reporting. This reflects the reporting of (OTC-)medication use in daily practice.

A limitation of this study is that, for the analyses, OTC-medication use was coded as present or absent and not subdivided per medication group. Explorative analyses of the predictors per subgroup of OTC-medication (e.g. analgesics, vitamins or medication for gastro-intestinal tract) showed that the predictors for each subgroup were comparable to the predictors of overall OTC-medication use (data available on request). However, a larger cohort is necessary to examine the strength of the associations per subgroup.

Another limitation of our study is that the prevalence of OTC-medication use may have been underestimated. The questionnaires have been filled out in different stages of pregnancy, with a range from 13 to 40 weeks. Especially self-reporting of (OTC-)medication later in pregnancy can lead to under-reporting due to poor recall. When a questionnaire is completed earlier in pregnancy the medication use afterwards is not known. Another potential cause of under-reporting is the questionnaire design, because an open-ended question was used to gain information about OTC-medication use. It has been reported that questions involving indication for use and drug-specific questions increase the prevalence estimates compared to open-ended questions. However, healthcare professionals commonly use open-ended questions. Therefore, we think that our data reflect the reporting in daily practice, while the actual use might be under-estimated. Healthcare professionals should take the questionnaire design into account while interpreting the results of this study. A third limitation is that the population is not representative for secondary healthcare. We assume that the predictors we have selected are still relevant in secondary healthcare, but other risk factors may contribute as well. Finally, the participants reported the number of GP visits during pregnancy one-month post-partum. This contradicts with the terms that only predictors that are available at the time the model will be used should be included in the prediction model [11]. Nevertheless we have decided to include this post-partum variable, because it is likely that there is an association between OTC-medication use during pregnancy and the number of GP visits. This association may exist, because pregnant women who visit their GP are more likely to suffer from pregnancy ailments and may use OTC-medication consequently.

A fourth limitation of our study is that the model was not internally validated. Therefore, we may have overestimated the predictive value of the model. Further research is recommended to validate the model and avoid the risk of overestimation.

Finally, we included five predictors in our model (nulliparity, prescription drug use, comorbidity, BMI26-30 kg/m² and GP visits) and found an area under the ROC-curve of 0.667. This value suggests that our model is moderately discriminating. On the other hand, the variability in predicted probabilities ranged from 6.6% to 57.4% and this indicates that the model is well discriminative. The Hosmer and Lemeshow test validated that the model is statistically fitted (P= 0.640).

To optimize the prediction model there is a need for studies with larger cohorts and studies that ascertain (OTC-)medication use more in detail. However, our study outcome gives a good indication of maternal characteristics that are positively associated with OTC-medication use and provides a risk score that is easy to implement in clinical practice.