Can we predict sleep-disordered breathing in pregnancy? The clinical utility of symptoms


  • Danielle L. Wilson,

    Corresponding author
    1. Institute for Breathing and Sleep, Austin Health, Heidelberg, Vic., Australia
    • Correspondence

      Danielle L. Wilson, Ground Floor Bowen Centre, Institute for Breathing and Sleep, Austin Health, Bowen Centre, Studley Road, Heidelberg, Vic. 3084, Australia. Tel.: 613-9496-3517; fax: 613-9496-5124; e-mail:

    Search for more papers by this author
  • Susan P. Walker,

    1. Department of Perinatal Medicine, Mercy Hospital for Women, Heidelberg, Vic., Australia
    2. Department of Obstetrics and Gynaecology, University of Melbourne, Parkville, Vic., Australia
    Search for more papers by this author
  • Alison M. Fung,

    1. Department of Perinatal Medicine, Mercy Hospital for Women, Heidelberg, Vic., Australia
    Search for more papers by this author
  • Fergal O'Donoghue,

    1. Institute for Breathing and Sleep, Austin Health, Heidelberg, Vic., Australia
    2. Department of Medicine, University of Melbourne, Parkville, Vic., Australia
    Search for more papers by this author
  • Maree Barnes,

    1. Institute for Breathing and Sleep, Austin Health, Heidelberg, Vic., Australia
    2. Department of Medicine, University of Melbourne, Parkville, Vic., Australia
    Search for more papers by this author
  • Mark Howard

    1. Institute for Breathing and Sleep, Austin Health, Heidelberg, Vic., Australia
    2. Department of Medicine, University of Melbourne, Parkville, Vic., Australia
    Search for more papers by this author


Sleep-disordered breathing (SDB) is reported commonly during pregnancy and is associated with an increased risk of adverse maternal and fetal outcomes, but the majority of these data are based upon self-report measures not validated for pregnancy. This study examined the predictive value of screening questionnaires for SDB administered at two time-points in pregnancy, and attempted to develop an ‘optimized predictive model’ for detecting SDB in pregnancy. A total of 380 women were recruited from an antenatal clinic in the second trimester of pregnancy. All participants completed the Berlin Questionnaire and the Multivariable Apnea Risk Index (MAP Index) at recruitment, with a subset of 43 women repeating the questionnaires at the time of polysomnography at 37 weeks' gestation. Fifteen of 43 (35%) women were confirmed to have a respiratory disturbance index (RDI) > 5 h−1. Prediction of an RDI > 5 h−1 was most accurate during the second trimester for both the Berlin Questionnaire (sensitivity 0.93, specificity 0.50, positive predictive value 0.50 and negative predictive value 0.93), and the MAP Index [area under the receiver operating characteristic (ROC) curve of 0.768]. A stepwise selection model identified snoring volume, a body mass index (BMI)≥32 kg m−2 and tiredness upon awakening as the strongest independent predictors of SDB during pregnancy; this model had an area under the ROC curve of 0.952. We conclude that existing clinical prediction models for SDB perform inadequately as a screening tool in pregnancy. The development of a highly predictive model from our data shows promise for a quick and easy screening tool to be validated for future use in pregnancy.


Self-reported symptoms of sleep-disordered breathing (SDB) such as snoring and witnessed obstructive apneas increase in frequency during pregnancy (Franklin et al., 2000; Pien et al., 2005); however, the accuracy of symptoms for detecting SDB in pregnancy is unclear. To date, there are no large studies that have confirmed a higher prevalence of SDB on polysomnography (PSG) among pregnant populations. The prevalence of SDB among pregnant women and the relationship of screening questionnaires to objectively confirmed SDB requires urgent clarification, given that preliminary studies suggest that SDB may increase the risk of adverse fetal and maternal outcomes, including hypertension, pre-eclampsia and impaired fetal growth (Bourjeily et al., 2010; Chen et al., 2011; Franklin et al., 2000).

Clinical prediction models based on self-reported symptoms of SDB, along with demographic and anthropometric data, have been developed to identify and prioritize those at highest risk of having SDB. A number of prediction models for SDB have shown potential utility in clinical settings; however, they have been validated mainly in predominantly male and/or middle-aged populations (Chung et al., 2008; Netzer et al., 1999; Takegami et al., 2009), usually with suspected sleep difficulties (Maislin et al., 1995). Only two studies have commented on the predictive capacity of screening questionnaires in pregnancy (Facco et al., 2012; Olivarez et al., 2010), concluding that the Berlin Questionnaire and Epworth Sleepiness Scale predict SDB poorly during pregnancy.

While studies using symptoms and clinical prediction models suggest an increased prevalence of SDB during pregnancy that is related to hypertensive disease (Bourjeily et al., 2010; Higgins et al., 2011; Olivarez et al., 2011), gestational diabetes mellitus (GDM) (Bourjeily et al., 2010; Reutrakul et al., 2011), delivery complications (Bourjeily et al., 2010) and low birth weight (Higgins et al., 2011; Olivarez et al., 2011), the accuracy of these methods has not been evaluated in this population. The predictive ability of clinical models needs to be documented in a pregnant population both for research and clinical purposes. This would enable more informed estimates of the prevalence of SDB during pregnancy and the relationship with adverse outcomes. Of great clinical relevance, an accurate screening tool would help identify pregnant women at risk and enable timely diagnosis of SDB during pregnancy with the potential to reduce adverse fetal and maternal outcomes.

Aims and hypotheses

The first aim of this study was to examine the predictive value of two questionnaires used commonly to identify patients with SDB, namely the Berlin Questionnaire (Netzer et al., 1999) and the Multivariable Apnea Risk Index (MAP Index) (Maislin et al., 1995), administered during the second and third trimesters of pregnancy. This study also aimed to identify the most significant risk factors on screening that predict SDB during pregnancy in order to develop an ‘optimized predictive model’ for SDB in pregnancy. The optimized model can be compared to existing models, and these data used to provide an estimate of SDB prevalence in a larger sample of pregnant women.


Study design and population

A prospective cohort study involving 380 women was conducted between May 2009 and November 2011. The Human Research Ethics Committees at Austin Health and Mercy Hospital for Women in Melbourne, Victoria, Australia approved the study and informed consent was obtained from all participants. Women in the second trimester of pregnancy were recruited from the antenatal clinics of the Mercy Hospital for Women. Exclusion criteria included multiple or complicated pregnancies, significant medical, psychological or psychiatric disorders diagnosed by a health professional, a previously diagnosed sleep disorder (e.g. SDB, insomnia, hypersomnolence) or current use of antidepressant medication.


All participants completed the Berlin Questionnaire and the MAP Index in the second trimester of pregnancy on the same day as initial recruitment into the study. Information was also gathered on age, parity, height, weight, neck circumference, gestational weight gain and pregnancy complications.

The Berlin Questionnaire (Netzer et al., 1999) asks about risk factors for sleep apnea. The questionnaire contains five questions concerning snoring, three questions address daytime sleepiness, and one question concerns history of high blood pressure. Predetermination of high risk and lower risk for sleep apnea was based on responses in three symptom categories. In category 1, high risk was defined as persistent symptoms (more than three to four times per week) in two or more questions about their snoring. In category 2, high risk was defined as persistent (more than three to four times per week) wake-time sleepiness, drowsy driving or both. In category 3, high risk was defined as a history of high blood pressure and/or a body mass index (BMI) more than 30 kg m−2. As per the questionnaire guidelines, to be considered as high risk for SDB a patient had to qualify as high risk for at least two symptom categories. Those with symptoms in at least two categories which were not persistent or who qualified in only one symptom category were considered low risk for SDB.

The MAP Index (Maislin et al., 1995) is based on a model developed for predicting the diagnosis of sleep apnea utilizing the self-reporting of loud snoring, snorting or gasping and breathing cessations. The frequency of each of the three symptoms over the past month was scored as follows: (0) = never, (1) = less than once a week, (2) = once or twice per week, (3) = three to four times per week, and (4) = five to seven times per week. An apnea symptom frequency index was calculated as the mean of the three apnea items, and therefore ranged from 0 to 4. The MAP Index score was developed using multiple logistic regression and incorporates the mean apnea symptom frequency along with age, BMI and gender; this score ranged from 0 to 1.

A subset of 120 participants was then invited to participate in a longitudinal study involving polysomnography (PSG) and repeat screening questionnaires at the end of pregnancy, 53 of whom agreed. This subset of participants were targeted as they scored at either end of the SDB risk spectrum (i.e. did not score as high risk in any Berlin Questionnaire category and reported ‘never’ experiencing SDB symptoms on the MAP Index, versus high risk responses in each of the Berlin Questionnaire categories and reporting frequent SDB symptoms on the MAP Index). Based on a previous study suggesting that the positive predictive value of the Berlin Questionnaire is only 19% (Olivarez et al., 2010), a larger proportion of high-risk women were targeted for the longitudinal study, with the aim of gathering data on similar numbers of women with and without SDB on PSG.


Overnight PSG was conducted at 37 weeks' gestation. In-laboratory studies were conducted to control for variations in potential external disruptions in the home environment, and to allow for accurate measurement of lights-out time. PSG was performed using the Somté (Compumedics, Abbotsford, Vic., Australia) portable sleep-monitoring device. Signals measured were electroencephalogram (EEG), electro-oculogram (EOG), electromyogram (EMG), nasal airflow measured via nasal cannula, arterial oxygen saturation, thoracic and abdominal respiratory effort via inductance plethysmography, snore, body position, leg movements and heart rate. PSG recordings were sleep-staged and respiratory scored by a single experienced sleep technologist who participates in internal and external quality assurance programmes, and who was blinded to SDB risk status, in accordance with current American Academy of Sleep Medicine (AASM) criteria, with the alternative definition used for scoring hypopneas (Iber et al., 2007). The number of apneas, hypopneas and respiratory effort-related arousals (RERAs) per hour of sleep was calculated and expressed as the respiratory disturbance index (RDI).

Statistical analysis

All statistical analyses were performed with spss version 17.0 (SPSS Inc., Chicago, IL, USA). Values are given in means with standard deviations [mean ± standard deviation (SD)] or median and interquartile range [median (IQR)] for non-normally distributed variables. SDB was defined as an RDI of more than five events per hour. A two-sided P-value of less than 0.05 was considered to indicate statistical significance.

Between-group comparisons were made using the chi-square test for independence for categorical variables, independent t-tests (two tailed) for normally distributed continuous variables and Mann–Whitney U-tests for non-normally distributed continuous variables.

Cross-tabulations were performed to determine sensitivity and specificity values for the Berlin Questionnaire when administered in the second and third trimesters of pregnancy. The predictive accuracy [sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV) and the likelihood ratio (LR)] of risk grouping and of each category was assessed for both RDI > 5 h−1 and RDI > 10 h−1. Receiver operator characteristic (ROC) curves were generated and the areas under the curves calculated to assess the predictive utility of the MAP Index.

Multivariate logistic regression was used to assess the relationship between SDB classified as a dichotomous variable (defined as an RDI > 5 h−1) and personal and symptom-related factors from the questionnaires, and odds ratios (ORs) were calculated. Stepwise logistic regression analysis was performed with the presence of SDB as the dependent variable, and explanatory variables with an α of less than 0.10 on univariate analysis were included. Due to very high collinearity between BMI and neck circumference, the latter redundant variable was removed from the model. Evaluation of the fit of the model showed that one case was considered an extreme outlier with a standardized residual (z) = 5.29; this case was removed from the model. The ROC curve and corresponding area under the curve was calculated to assess the predictive ability of this model.

In order to estimate prevalence of SDB in the questionnaire sample, we calculated the positive and negative predictive values for high risk on the new multivariate logistic regression model for detecting SDB using data from the PSG sample. These values, along with the prevalence of a high risk score on the new model, were used to estimate the prevalence of SDB in the whole questionnaire sample using the method described by Marshall (1990).



We recruited 380 women to the study, of whom 372 completed the second-trimester questionnaires. Of the 120 women invited to participate in the longitudinal study, 53 women agreed and repeated the screening questionnaire at the time of the PSG at 37 weeks' gestation. Seven of these participants withdrew from the study later and three delivered prior to 37 weeks, leaving 43 participants for analysis.

Women who completed the PSG were more likely to be older and had a higher BMI. The clinical characteristics of those completing the questionnaires compared to those who underwent PSG are summarized in Table 1.

Table 1. Characteristics of polysomnography and questionnaire samples
 Polysomnography (n = 43)Questionnaire (n = 329) P
  1. BMI, body mass index; GDM, gestational diabetes mellitus; PSG, polysomnography.

Age (years)33.5 ± 5.130.8 ± 5.30.002
Gestation at recruitment (weeks)22.3 ± 4.021.3 ± 2.20.15
Nulliparous (%)11 (25.6)143 (43.5)0.022
BMI pre-pregnancy (kg m−2)32.2 ± 8.025.3 ± 5.6<0.001
BMI at recruitment (kg m−2)34.5 ± 7.527.6 ± 5.6<0.001
BMI at PSG (kg m−2)37.5 ± 7.9
Neck circumference (cm)36.2 ± 3.233.6 ± 2.4<0.001
GDM8 (19%)43 (13.3%)0.33
Gestational hypertension8 (19%)23 (7.1%)0.02
Pre-eclampsia2 (4.8%)15 (4.6%)0.97
Smoker5 (11.6%)39 (11.9%)0.95
Asthma7 (16.3%)31 (9.4%)0.32


Sleep-disordered breathing (RDI > 5 h−1) was found in 15 of the 43 women who underwent PSG (35%), and an RDI > 10 h−1 was found in seven (16%); the median (IQR) for RDI = 3.5 (1.8–6.6). Sleep stage architecture did not differ between those with an RDI > 5 h−1 and those with an RDI ≤ 5 h−1. The RDI > 5 h−1 group had a higher apnea/hypopnea index [AHI; 6.2 (4.9–13.2) versus 1.5 (0.6–2.7), P < 0.001], more oxygen desaturations of at least 4% per hour of sleep [3.9 (1.1–8.8) versus 0.4 (0.2–1.1), P < 0.001], and spent a higher proportion of the night snoring [24.9% (7.5–51.1) versus 6.1% (0.2–17.0), P = 0.03] compared to those without SDB.

The Berlin Questionnaire

The number of participants at elevated risk in each category of the Berlin Questionnaire and those qualifying as high risk for SDB on the overall Berlin score are shown in Table 2. Overall high risk for SDB was significantly more common in the PSG sample compared to the questionnaire sample, and the number of women at high risk in each category increased from the second trimester to 37 weeks when the PSG was recorded. For overall risk specifically, one woman was high risk in the second trimester but became low risk at 37 weeks, whereas five women were low risk in the second trimester then became high risk at 37 weeks.

Table 2. Berlin Questionnaire and MAP Index results for polysomnography and questionnaire samples
 Questionnaire sample second trimester (n = 329)PSG sample second trimester (n = 43)PSG sample 37 weeks (n = 43) P
  1. P-value given for chi-square or independent t-test between questionnaire sample and PSG sample second trimester.

Berlin high risk80 (24.5%)28 (65.1%)32 (74.4%)<0.001
Category 1 high risk82 (25.0%)28 (65.1%)30 (69.8%)<0.001
Category 2 high risk112 (34.1%)21 (48.8%)26 (60.5%)0.063
Category 3 high risk102 (31.0%)32 (74.4%)37 (86.0%)<0.001
MAP Index score0.10 ± 0.100.25 ± 0.190.34 ± 0.22<0.001
MAP Index apnea symptom frequency index0.24 ± 0.530.95 ± 1.141.10 ± 1.22<0.001

Predictive value of the Berlin Questionnaire for SDB on PSG

Those at high risk of SDB on the Berlin Questionnaire at the time of the PSG had significantly higher RDI per hour scores [4.6 (3.3–7.4) versus 1.5 (0.9–1.9); z = −2.81, P = 0.004] compared to those at low risk.

Table 3 shows the ability of risk grouping on the Berlin Questionnaire to predict an RDI > 5 h−1 and an RDI > 10 h−1. The questionnaire showed better predictive value in the second trimester than the third trimester. Nevertheless, while the sensitivity of the questionnaire was high with a correspondingly high NPV, the specificity and PPV was poor in both trimesters. Of the 74% of women in our PSG sample classified as high risk, fewer than half had confirmed SDB. The predictive ability of the Berlin Questionnaire was even poorer for identifying those with an RDI of >10 h−1.

Table 3. Diagnostic test thresholds for the Berlin Questionnaire in the second and third trimesters of pregnancy
 SensitivitySpecificityLikelihood ratio (relative risk)PPVNPV
  1. PPV, positive predictive value; NPV, negative predictive value; T2, second trimester; T3, third trimester; RDI h−1, Respiratory Disturbance Index per hour.

RDI > 5 h−10.930.501.870.500.93
RDI > 10 h−10.860.391.400.210.93
RDI > 5 h−10.870.321.280.410.82
RDI > 10 h−10.860.

The MAP Index

The results for the MAP Index score and the apnea symptom frequency index for the PSG and Questionnaire samples are shown in Table 2. The MAP Index score increased significantly between the second trimester and 37 weeks' gestation for the PSG sample, t(42) = 6.55, P < 0.001, whereas the apnea symptom frequency index did not, t(42) = 1.04, P = 0.31. This discrepancy is due probably to the incorporation of increasing BMI into the MAP Index score, but not the apnea symptom frequency index.

Predictive value of the MAP Index for SDB on PSG

Figure 1 presents ROC curves for the predicted probabilities derived from the MAP index and the apnea symptom frequency index at the second and third trimesters of pregnancy, for SDB defined as an RDI > 5. The apnea symptom frequency index was more accurate at predicting the presence of SDB in this population than the complete MAP Index which includes gender, age and BMI. This was consistent for questionnaire responses from both the second and third trimesters, but this improvement in predictive values was not significant (P = 0.68 and P = 0.52, respectively). The MAP Index administered during the second trimester compared to the third, however, was significantly more predictive of SDB in late pregnancy (AUC = 0.733 versus AUC = 0.643, = 0.02; see Fig. 1a,b).

Figure 1.

Receiver operating characteristics (ROC) curves for the predicted probabilities derived from the Multivariable Apnea Risk Index and apnea symptom frequency score during the second (T2; a) and third (T3; b) trimester of pregnancy for sleep-disordered breathing (SDB) as defined as an respiratory disturbance index (RDI) > 5. Also displayed is the ROC curve for the new optimized model based on the stepwise logistic regression for predictors of SDB during pregnancy (c).

Using only the apnea symptom frequency index from the MAP Index (which ranges from 0 to 4) during the second trimester had the highest predictive ability. Cut-points were determined from the ROC curve to identify those with SDB with a high sensitivity and alternatively a cut-point with a high specificity. Choosing a cut-off score of 0.5 achieves a sensitivity of 80%, and with a specificity of 71% gives a likelihood ratio of 2.8, a PPV of 60% and a NPV of 87%. Choosing a cut-off score of 1.84 achieves a high specificity of 89%, with a sensitivity of 53%, an LR of 5.0, a PPV of 73% and an NPV of 78%.

Optimized prediction of SDB during pregnancy

When all clinical predictors were subject to multivariate analysis, the factors associated with an increased risk of SDB during pregnancy were a BMI ≥ 32 kg m−2 when the questionnaire was completed at recruitment, an increase in neck circumference, history of asthma, volume of reported snoring, frequency of reported snoring, breathing pauses, tiredness upon awakening, daytime tiredness and self-reported high blood pressure (Table 4).

Table 4. Demographic and questionnaire-based factors at the second trimester associated with sleep-disordered breathing (SDB) during pregnancy
VariableUnits or category (range/SD)OR (95% CI)P-value
  1. OR, odds ratio; CI, confidence interval; SD, standard deviation; BMI, body mass index; GH, gestational hypertension; GDM, gestational diabetes mellitus.

  2. Gestational weight gain is from pre-pregnancy until recruitment into study (mean = 22 weeks).

  3. a

    OR for continuous and ordinal variables indicate the change in odds for an increase of 1 standard deviation or one category, respectively.

  4. b

    Categories for volume of snoring were (0) no snoring, (1) slightly louder than breathing, (2) as loud as talking, (3) louder than talking, (4) very loud.

  5. c

    Categories for frequency were (1) never or nearly never, (2) 1–2 times per month, (3) 1–2 times per week, (4) 3–4 times per week, (5) nearly every day.

Agea (years)22–41/5.21.64 (0.81–3.33)0.17
ParityTimes given birth (0–3)0.56 (0.22–1.42)0.22
Pre-pregnancy BMIakg m−2 (20.6–56.8/8.0)1.72 (0.85–3.28)0.12
Recruitment BMIakg m−2 (20.6–56.8/7.5)1.64 (0.86–3.25)0.16
Gestational weight gainakg (−9.3 to 27.0/5.8)0.85 (0.46–1.56)0.61
BMI ≥ 32 kg m−2Yes/no15.00 (1.72–130.76)0.01
Neck circumferencea(cm)29–43/3.22.32 (1.07–5.06)0.04
GH or pre-eclampsiaYes/no1.33 (0.31–5.73)0.70
GDMYes/no4.00 (0.80–20.02)0.09
AsthmaYes/no6.50 (1.08–39.11)0.04
SmokerYes/no1.39 (0.20–9.45)0.74
SnoringYes/no5.60 (0.63–49.94)0.12
Snoring volumeaVolume (0–4)b2.19 (1.23–3.90)0.008
Snoring frequencyaFrequency (1–5)c1.96 (1.12–3.42)0.02
Snorting/gaspingaFrequency (1–5)c1.32 (0.90–1.95)0.16
Breathing pausesaFrequency (1–5)c2.25 (1.21–4.15)0.01
Tired upon awakeningaFrequency (1–5)c2.50 (1.23–5.07)0.01
Daytime tiredness/fatigueaFrequency (1–5)c3.50 (1.34–9.12)0.01
Reported hypertensionYes/no8.67 (1.48–50.92)0.02

Using the data available at the time when the questionnaires were completed in the second trimester, a stepwise selection model identified snoring volume, a BMI ≥ 32 kg m−2 and tiredness upon awakening as the strongest independent predictors of presence of SDB on the sleep study (χ2 = 31.19, ≤ 0.001, R2 = 0.61; Table 5). The ROC curve for the predicted probabilities of the optimized predictive model for SDB (defined as an RDI > 5 h−1) is shown in Fig. 1c, with an area under the curve of 0.952. The predictive characteristics of this optimized model are shown in Table 6.

Table 5. Factors associated with SDB during pregnancy on stepwise logistic regression model
VariableCoefficientOR (95% CI)P-value
  1. OR, odds ratio; BMI, body mass index; CI, confidence interval.

  2. a

    OR for ordinal variables indicate the change in odds for an increase of one category.

  3. b

    Categories for volume of snoring were (0) no snoring, (1) slightly louder than breathing, (2) as loud as talking, (3) louder than talking, (4) very loud.

  4. c

    Categories for frequency were (1) never or nearly never, (2) one to two times per month, (3) one to two times per week, (4) three to four times per week, (5) nearly every day.

Snoring volumea2.4011.00 (1.65–73.20)b0.01
BMI ≥ 32 kg m−24.62101.36 (1.79–5733.37)0.03
Tired upon awakeninga2.239.26 (1.71–50.07)c0.01
Constant−17.921.65e-08 (5.04e-14–0.01)0.68
Table 6. Characteristics of the prediction equation for an RDI > 5 hr−1 on the optimised prediction model
ValueSensitivity (%)Specificity (%)PPV (%)NPV (%)
  1. PPV, positive predictive value; NPV, negative predictive value.


An example of the utilization of this model in questionnaire form is given in the Appendix. Within the PSG sample, this questionnaire has the same predictive properties of when the optimized predictive model is used with a cut-off value of 0.30, with a sensitivity of 85%, a specificity of 96%, a PPV of 92% and an NPV of 93%.

Prevalence estimate of SDB in questionnaire sample

Based on the method described by Marshall (1990), an estimate of the true proportion of SDB prevalence in the large questionnaire sample can be made from the PSG population using the new prediction model with known error. First, we calculated that the PPV and NPV for high risk on the new optimized prediction model from the PSG sample data using a cut-off score of 0.30 (see Table 6) is 92 and 93%, respectively. Based on this cut-off score of 0.30, we determined that 4.5% of the questionnaire sample was at high risk for SDB. Using the equation below, we estimate a prevalence of 10.8% for at least an RDI > 5 in the questionnaire sample: PPV × measured proportion with disease + (1−NPV) × measured proportion without disease = estimated true proportion with disease 0.92 × 0.045 + [(1−0.93) × 0.955] = 0.10825 = 10.8%.


In this longitudinal prospective study the Berlin Questionnaire and MAP Index had low to moderate predictive capacity in pregnancy. A model including snoring volume, obesity and morning tiredness provided greater diagnostic accuracy for SDB in this population, but requires prospective validation. The two existing prediction models for SDB will typically overestimate the true prevalence in pregnancy. High sensitivity allows the questionnaires to detect those who do have SDB; however, low specificity and poor PPVs mean that false positive diagnoses would be common. For example, of the 108 (29%) women in our total cohort classified as high risk on the Berlin Questionnaire, we would expect that half would be cleared subsequently of SDB following PSG. On a positive note, the MAP index has the capacity for improved specificity and PPV by redefining the ROC curve cut-point as demonstrated, but in turn poor sensitivity would result in only half of those with the disease being identified. Therefore, while self-report measures are administered easily and allow for the collection of large amounts of data, they are inadequate for the diagnosis of SDB during pregnancy without confirmatory objective testing. Utilizing questionnaires alone is likely to result in significant inaccuracy in SDB prevalence estimates and, more importantly, assumed relationships between SDB and maternal and fetal outcomes based purely on reported symptoms may be imprecise.

The predictive value of screening questionnaires was no worse, if not marginally better in the second rather than third trimesters in this study. General sleep disturbance and fatigue is more common with advancing gestation (Wilson et al., 2011), and we postulate that these increasing symptoms confound the association between the questionnaires and SDB diagnosis. While the second-trimester values for the Berlin Questionnaire remain too low to be of real clinical value, the predictive values for the MAP Index, particularly during the second trimester, may give it potential utility as a preliminary screening device due to its ability to prioritize sensitivity or specificity by adjusting cut-points on the ROC curve. Despite their flaws, the fact that these screening tools have at least comparable predictive value in the second trimester is encouraging for future developments, given that earlier detection of disease would allow more time for effective intervention.

Using our data, we were able to develop an optimized model with very high predictive properties for screening of SDB during pregnancy, to enable better identification of patients most likely to have the disease. The model identified three factors, obesity (BMI ≥ 32 kg m−2), snoring volume and tiredness upon awakening, to be the strongest independent predictors of SDB. Interestingly, univariate analysis showed that a diagnosis of gestational hypertension or pre-eclampsia by the end of pregnancy was not associated significantly with SDB, but self-reported hypertension by the second trimester was. The recognition of hypertension as an important risk factor for SDB is not surprising, given the recent abundance of literature suggesting a link between women with symptoms of SDB and the development of hypertensive disorders (Bourjeily et al., 2010; Franklin et al., 2000). This optimized model now requires application to a larger cohort in order to be validated.

Why are the existing predictive questionnaires for SDB not so successful for pregnant women? The key problem was that they overestimated the likelihood of objectively measured SDB based on symptoms. While pregnancy-related symptoms often associated with SDB such as increased daytime fatigue, nasal hyperaemia and respiratory-related changes such as upper airway oedema may contribute, we postulate that the treatment of the BMI data may be a factor. First, a pregnant woman is more likely to have an increased BMI while carrying a lesser proportion of body fat and is therefore more likely to meet the criteria of a BMI ≥ 30 kg m−2 used by the Berlin Questionnaire. Secondly, excess weight is a well-established predictor of SDB (Young et al., 2005); however, within our data both pre-pregnancy BMI and BMI at the time of recruitment into the study were not associated with an RDI > 5. The predictive ability of the MAP Index was consequently improved with the removal of BMI from the model. However, there appears to be a non-linear relationship between BMI and RDI per hour in our data; none of the women with a BMI of <32 kg m−2 had an RDI > 5 h−1, whereas for those with a BMI of >32 kg m−2 there appeared to be no relationship at all between BMI and RDI per hour. To illustrate, of the nine women with a BMI > 40 kg m−2 when the questionnaire was completed in the second trimester, only a third were diagnosed with mild-to-moderate SDB. Unlike previous similar studies by Olivarez et al. (2010) and Facco et al. (2012), which found that BMI as a continuous measurement was predictive of SDB in pregnancy, we found that BMI needed to be treated as a dichotomous variable, with a higher cut-off than that used by the Berlin Questionnaire.

Historically, SDB has been described as a primarily male disease, and as such prediction models have typically been developed using middle-aged populations with a male predominance. As such, the factors of age and gender included in the MAP Index are generally not useful for women of child-bearing age. A study by Rowley et al. (2000) compared four clinical prediction models for SDB and found that all models performed better for men than for women. Investigation into gender differences in the clinical manifestation of SDB have found that women with SDB are likely to describe their sleepiness differently to men (Chervin, 2000), and are less likely than men to report classic symptoms of SDB such as snoring, gasping, snorting and apnea (Redline et al., 1994). Furthermore, studies of clinical samples have found that women with SDB are more likely to report symptoms such as daytime tiredness, morning headache, insomnia and mood disturbance (Quintana-Gallego et al., 2004; Shepertycky et al., 2005). Questions regarding overt SDB symptoms may not be as appropriate for identifying SDB in females, particularly during pregnancy. Our research suggests that physical factors (obesity and neck circumference), subtle symptoms of SDB and hypertension are important predictors of the presence of the disease pregnant women.

The potential to develop a brief validated questionnaire as a predictive clinical tool is encouraging. Prenatal health-care professionals could easily perform this quick and inexpensive survey, and then refer the high-risk pregnant women immediately for a diagnostic sleep study rather than awaiting full clinical evaluation by a sleep specialist. This, in turn, should allow at-risk pregnant women to advance onto treatment in a more timely manner if needed.

The prevalence of SDB in women of childbearing age has been found to be as low as 0.6% (Bixler et al., 2001) or as high as 10.8% (Young et al., 2003). Application of our optimized prediction model to our questionnaire cohort data estimates the prevalence of SDB (defined as an RDI h−1 > 5) during pregnancy to be 10.8%. Questionnaire-based estimates of SDB prevalence during pregnancy are often higher, ranging from 29 to 33% (Higgins et al., 2011; Reutrakul et al., 2011), and likely to be overestimated due to the high false positive rate. Utilizing this more accurate model developed within a pregnant population in conjunction with methods to adjust for the false positive and negative rates provides a more accurate estimate of the prevalence of SDB in this population (Marshall, 1990). Given the small sample size used to arrive at this estimate we acknowledge the potential for error, but we believe these data add further insight into the prevalence of SDB during pregnancy, which remains unknown due to the paucity of objective data.

Strengths and limitations

An important strength of this study was the validation of SDB diagnosis using full overnight PSG, enabling the measurement of both discrete and subtle respiratory events, including those causing cortical arousal disrupting sleep. Abbreviated devices allow estimation of AHI based on limited data, but only full PSG can detect all respiratory events indexed per hour of sleep. However, using full PSG limited the sample size recruited to the study, and subsequently only a relatively small number were diagnosed with SDB. As mentioned, the study aimed to collect PSG data on equal numbers of women with and without SDB, but the unexpectedly poor sensitivity of the questionnaires resulted in many false positives. Prospective validation of the new prediction model on a larger sample is required, and is planned in future studies.

Our data suggest that although the predictive values of screening questionnaires in pregnancy are largely inadequate, they are at least comparable in the second compared to the third trimesters. As we measured SDB objectively at 37 weeks' gestation, we can only refer to the prediction of SDB at that time and not whether or not it was present during the second trimester. Due to the lack of longitudinal data on the pathogenesis of SDB during pregnancy, we cannot say whether SDB diagnosed towards the end of pregnancy was present throughout or only emerged at some later stage during gestation. Future studies need to address this; if interventional studies propose to treat SDB in pregnancy, the diagnosis needs to be made early enough to allow time for this intervention to be effective in improving maternal and fetal outcomes.

This study could be criticized for defining SDB as a relatively low RDI of >5 h−1, but there are arguments for using this criterion. Subtle airflow abnormalities such as nasal flow limitation and RERAs are common in pregnancy (Guilleminault et al., 2000b), and even non-pregnant women are more likely to display this pattern of SDB rather than discrete episodes of obstruction and oxygen desaturation characteristic of sleep apnea (Guilleminault et al., 2000a). As a consequence, women are often observed to have a lower AHI than men (O'Connor et al., 2000) and are more symptomatic at a lower disease severity level (Young et al., 1996). Even low levels of SDB has been shown to have haemodynamic consequences with recent data suggesting that treatment of mild SDB with continuous positive airway pressure (CPAP) is associated with improvements in blood pressure in pre-eclamptic patients (Edwards et al., 2000; Poyares et al., 2007). We believe that defining SDB by an RDI of >5 h−1 in pregnancy is justified, although the relationship with important fetal and maternal outcomes remains to be determined.


Clinical prediction models for SDB that are used commonly in other populations are poor at identifying SDB in pregnancy, with high false positive rates. This would tend to dilute the relationship between SDB and pregnancy outcomes identified using these methods and demonstrates the need to use objective methods to assess the relationship between SDB and pregnancy outcomes. Clinically, beyond using the MAP index as a potential preliminary screening tool, pregnant women with a suspicion of SDB should be taken care of individually on a case-by-case basis.

The high predictive accuracy of the three-item regression model developed from our data shows promise that a quick and easy screening tool for SDB in pregnancy can be validated for future use, but requires prospective evaluation. It is also encouraging that the MAP Index administered in the second trimester may be capable of screening for SDB in the third trimester within our cohort, as the potential for earlier screening with the development of more discriminative tools will hopefully allow earlier detection of disease and more time for effective intervention and potentially improved perinatal outcomes.

Conflict of Interest

No conflicts of interest declared.


The authors would like to thank the staff at the Austin Health Sleep Laboratory and the antenatal clinics at Mercy Hospital for Women for their support of this project; we appreciate the valuable contribution made by each of the research participants. We would also like to thank Dr Helen Esdale and Ms Gabrielle Fleming for their hard work in recruiting for this study. This study was supported by the Austin Medical Research Foundation, the Australian Stillbirth Alliance, and the Medical Research Foundation for Women and Babies.