• Epidemiology;
  • pregnancy risk factors;
  • prevention;
  • spontaneous abortion


  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Conclusions
  8. References


To identify modifiable risk factors for miscarriage and to estimate the preventable proportion of miscarriages that could be attributed to these.


Nationwide observational follow-up study.




Ninety-one thousand four hundred and twenty seven pregnancies included in the Danish National Birth Cohort between 1996 and 2002.


Information on potentially modifiable risk factors before and during pregnancy was collected by means of computer-assisted telephone interviews and linkage with Danish registers, ensuring almost complete follow-up of pregnancy outcome. Modifiable risk factors for miscarriage were identified by multiple Cox regression analysis, which provided the background for our estimations of population attributable fractions. In all, 88 373 pregnancies had full information on all covariates and were included in this analysis.

Main outcome measures

Miscarriage before 22 completed weeks of gestation.


The potentially modifiable pre-pregnant risk factors associated with increased miscarriage risk were: age of 30 years or more at conception, underweight, and obesity. During pregnancy the modifiable risk factors were: alcohol consumption, lifting of >20 kg daily, and night work. We estimated that 25.2% of the miscarriages might be prevented by reduction of all these risk factors to low risk levels. Modification of risk factors acting before and during pregnancy could lead to prevention of 14.7 and 12.5%, respectively, of the miscarriages. Maternal age at conception and alcohol consumption were the most important risk factors.


Miscarriage risk is increased by multiple potentially modifiable risk factors and a considerable proportion of miscarriages may be preventable.


  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Conclusions
  8. References

Miscarriage is the most common adverse pregnancy outcome. As miscarriages are considered irreversible, prevention is probably the only way to intervene in this problem, which has detrimental psychological consequences for the woman and her partner[1] and delays successful childbearing. Chromosomal abnormality of the fetus is involved in approximately half of the miscarriages, for which errors at meiosis caused by advanced maternal age at time of conception has been found to be a risk factor.[2, 3] Several other potentially modifiable risk factors during pregnancy have been examined, e.g. alcohol consumption,[4-7] caffeine consumption[8, 9] and smoking[5, 10-13]; however, the results have been inconsistent and have primarily been evaluated on the relative scale. Studies of multiple risk factors for miscarriage were identified.[14] A few studies have also presented the proportions of miscarriages that might be attributed to the risk factors examined, which primarily have been smoking and alcohol.[5, 12, 13, 15] Studies presenting preventable proportions of miscarriages related to multiple potentially modifiable risk factors are lacking although they may be useful as guidance for future prevention strategies towards miscarriage.

The Danish National Birth Cohort (DNBC), which has more than 100 000 pregnancies and several advantages regarding miscarriage research,[16] was used for this study. We aimed to identify potentially modifiable risk factors for miscarriage in the DNBC and to estimate the preventable proportions of miscarriages that are related to these factors.


  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Conclusions
  8. References

Study design

We conducted a follow-up study based on pregnancies enrolled in the DNBC between 1996 and 2002.[17] The recruitment of participants occurred in connection with the first antenatal visit at the general practitioner, which in Denmark normally takes place shortly after recognition of the pregnancy. The mean gestational age at enrolment to study was 10.6 (SD 3.4) completed weeks. The pregnant women provided informed consent, and permission from the Danish Data Protection Board and the Scientific Ethics Committees was obtained before initiation of the study.

A computer-assisted telephone interview was used to collect information on lifestyle-related risk factors before and during the first part of the pregnancy, reproductive history and occupation. The interview was planned to take place at gestational week 12 or as soon as possible after that. Women who experienced a miscarriage prior to the scheduled interview were offered an interview with similar questions.[17] The questionnaire used can be found at

Study population

We included all pregnancies with information on risk factors for miscarriage that was provided either in an interview during pregnancy or after a miscarriage. Pregnancies with no interview were excluded. We set some further exclusion criteria: hydatidiform mole and ectopic pregnancies, pregnancies enrolled in the DNBC at 22 gestational weeks or later, and pregnancies with no information on gestational age. For the multiple Cox regression analysis, pregnancies with missing values on risk factors or confounders were excluded (Figure 1).


Figure 1. Flowchart of pregnancies from enrolment in the Danish National Birth Cohort (DNBC) to pregnancies eligible for multiple Cox regression analysis.

Download figure to PowerPoint

Ascertainment of miscarriage

Miscarriage, in Denmark defined as a spontaneous fetal death and/or expulsion before 22 completed weeks of gestation, was the outcome of interest in this study. Information on pregnancy outcome was obtained from the Danish registers and the women's self-reported information. The Civil Registration System was used to obtain information on date of birth for live born children and the mother's emigration or death during pregnancy. The National Patient Register contains discharge diagnoses according to ICD-10 for all hospitalisations and was used to obtain information on other pregnancy outcomes (i.e., ectopic pregnancy, induced abortion, hydatidiform mole, miscarriage and stillbirth). For non-hospitalised miscarriages the outcome status was traced from the interview data.

Ascertainment of risk factors

A priori, we reviewed the literature of risk factors for miscarriage and selected nine established risk factors for analysis based on the notion that they were potentially modifiable, i.e., at least to some extent within the women's or legislative control, and possible to evaluate due to their measurement in the DNBC study. We distinguished between risk factors that may be modifiable during pregnancy and pre-pregnant risk factors that in principle are modifiable, but would probably require more profound structural and cultural changes. Continuous information on risk factors during pregnancy were categorised as: amount of exercise (0 minutes/week, 1–60 minutes/week, 61–120 minutes/week, 121–180 minutes/week, 181–300 minutes/week, >300 minutes/week), alcohol consumption (0 drinks/week, ½–1½ drinks/week, 2–3½ drinks/week, 4+ drinks/week), smoking (0 cigarettes/day, 1–10 cigarettes/day, 11+ cigarettes/day), and coffee consumption (0 cups/day, ½–7½ cups/day, 8+ cups/day). The following categorical variables were grouped as: work schedule (not working, daytime, fixed evening, fixed night, rotating shift [without night], rotating shift [including night]), and lifting of >20 kg daily (no lifting of >20 kg daily, lifting of >20 kg daily). Risk factors before pregnancy were defined as: maternal age at conception (<25 years, 25–29 years, 30–34 years, 35–39 years, 40+ years), pre-pregnant weight status (underweight ‘<18.5 kg/m2’, normal weight ‘18.5–24.9 kg/m2’, overweight ‘25.0–29.9 kg/m2’, obese ‘≥30 kg/m2’, and previously diagnosed genital diseases (no diagnosis of genital disease, ever had cervical conisation, dysplasia in the cervix, other genital diseases).

Statistical analysis

We constructed a Cox proportional hazards model including the nine above-mentioned potentially modifiable risk factors. In addition, parity and highest household occupational status were included as confounders; we judged these to be non-modifiable. Paternal age at conception was not included, as it was unrelated to miscarriage risk after adjusting for maternal age,[18] and as this variable would reduce the number of pregnancies substantially due to a high number of missing values on this variable. Previous miscarriage was potentially a confounder, but adjustment might introduce bias.[19] We tested whether inclusion of this variable changed the risk estimates of the other variables in the model. Gestational age (i.e., days since first day of last menstrual period) was the time variable. We used the Cox regression model as this allowed us to take into account the varying times of recruitment among women, the time-varying effects of covariates and incomplete follow-up. A pregnancy was observed to be at risk of miscarriage from the enrolment to the study and until the date of the first of the following events: miscarriage, induced abortion on request, loss to follow-up, or survival of at least 22 weeks of gestation. As some of the women contributed to the study with more than one pregnancy, we used robust standard errors to take into account the potential dependency between some of the pregnancies.[20] All analyses were performed using the SAS software version 9.2 (SAS Institute, Cary, NC, USA).

From a clinical and prevention point of view, it is relevant to distinguish between early and late abortions; thus, separate effects of each covariate on miscarriage were analysed for the first and second trimester of pregnancy, i.e., defined as <12 and 12+ completed gestational weeks. This separation of trimesters was also used to examine the assumption of proportional hazards for all covariates in the regression model; differences in effects on miscarriage between trimesters were tested with a standard Wald test.

As women who report on risk factors after having experienced a negative pregnancy outcome often have better recall than others or deny certain risk factors, it is possible that associations in this study can be explained by recall bias.[21] Thus, the multiple Cox regression analysis was carried out for a sub-cohort restricted to pregnancies for which information on risk factors was collected while the women were still pregnant to elucidate whether the estimates for miscarriage risk were affected by recall bias. We were able to conduct this analysis for the second trimester of pregnancy only.

The experience of an earlier successful or unsuccessful pregnancy is likely to modify a woman's risk behaviour.[22] Consequently, an analysis restricted to primigravid women who became pregnant within 0–5 months of trying was conducted to examine whether the woman's behaviour during pregnancy was affected by her reproductive history, i.e., behaviour modification bias.

The risk of miscarriage for each individual woman was estimated from the Cox regression model and provided the background for estimating the population attributable fraction (PAF). The PAF was defined as the proportionate excess risk of miscarriage associated with exposure to a risk factor and was estimated using a formula identical to the one introduced by Levin in 1953: inline image.[23, 24] However, in our situation pT was estimated as the average of each woman's risk of miscarriage obtained from the multiple Cox regression model, and the probability p0 was estimated similarly[24]; however, instead of estimating each woman's individual risk based on her observed risk factors, prevention scenarios were defined as low risk levels of relevant risk factors, i.e., all women's alcohol consumption during pregnancy was set to zero in the Cox regression model. We had limited information on pregnancies before six gestational weeks; consequently, PAF was estimated conditional on the pregnancy lasting for at least 6 weeks. As pregnancies ending in induced abortions introduce a risk competing with that of miscarriage, we tested whether the results of a model in which competing events were censored, differed from a model where induced abortions were considered as an event together with miscarriage.


  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Conclusions
  8. References

The study population consisted of 91 427 pregnancies, of which 3177 resulted in a miscarriage (Figure 1); as the mean gestational age at interview was 16 weeks, a considerable proportion of the women who experienced a miscarriage were interviewed after the miscarriage had occurred (n = 2444; Figure 2). Table 1 presents characteristics of the women whose pregnancies were included in the cohort as well as the mean gestational age at enrolment in the study. According to the table, the assumption that data are missing completely at random according to the relevant risk factors and confounders is not contradicted.

Table 1. Characteristics of and time of enrolment in study for 91 427 pregnancies in the Danish National Birth Cohort, 1996–2002
 Pregnancies (= 91 427) n (%)bMiscarriages (= 3177) n (%)bGestational age at enrolmenta Mean (SD)
  1. a

    Measured in completed gestational weeks.

  2. b

    Percentages do not add up to 100 in some cases because of rounding.

  3. c

    No information was available if the woman did not know what to answer, did not wish to answer, or if the interviewer categorised the woman's answer as undefined.

Reproductive characteristics
Previous miscarriage
073 753 (80.7)2430 (76.5)10.5 (3.4)
113 265 (14.5)528 (16.6)10.6 (3.4)
23083 (3.4)131 (4.1)10.9 (3.4)
3+1248 (1.4)82 (2.6)11.1 (3.5)
Missingc78 (0.1)6 (0.2)10.2 (3.5)
Time to pregnancy (months)
Not planned10 300 (11.3)337 (10.6)11.2 (3.7)
0–556 156 (61.4)2011 (63.3)10.4 (3.4)
6–1212 562 (13.7)439 (13.8)10.4 (3.3)
>1212 104 (13.2)381 (12.0)10.8 (3.3)
Missingc305 (0.3)9 (0.3)11.5 (3.6)
Nulliparous43 088 (47.1)1407 (44.3)10.4 (3.4)
Parous48 282 (52.8)1768 (55.6)10.7 (3.4)
Missingc57 (0.1)2 (0.1)9.8 (3.4)
Highest household occupational status
Higher grade professionals21 698 (23.7)817 (25.7)10.5 (3.4)
Lower grade professionals28 413 (31.1)1001 (31.5)10.5 (3.4)
Skilled workers25 436 (27.8)789 (24.8)10.7 (3.4)
Unskilled workers12 821 (14.0)457 (14.4)10.6 (3.5)
Students2146 (2.3)82 (2.6)10.1 (3.5)
Economically inactive >1 year708 (0.8)29 (0.9)10.9 (4.0)
Missingc205 (0.2)2 (0.1)11.0 (3.4)
Potentially modifiable risk factors
Maternal age at conception (years)
<2511 689 (12.8)355 (11.2)10.5 (3.5)
25–2937 769 (41.3)1153 (36.3)10.4 (3.4)
30–3431 182 (34.1)1108 (34.9)10.7 (3.4)
35–399813 (10.7)468 (14.7)10.8 (3.4)
40+974 (1.1)93 (2.9)11.1 (3.6)
Missingc0 (0.0)0 (0.0)
Amount of exercise during pregnancy (minutes/week)
057 592 (63.0)1599 (50.3)10.7 (3.4)
1–6012 099 (13.2)371 (11.7)10.6 (3.4)
61–1209220 (10.1)425 (13.4)10.4 (3.4)
121–1805086 (5.6)286 (9.0)10.3 (3.3)
181–3004619 (5.1)268 (8.4)10.2 (3.3)
>3002602 (2.8)218 (6.9)10.3 (3.4)
Missingc209 (0.2)10 (0.3)10.9 (3.7)
Alcohol consumption during pregnancy (drinks per week)
050 553 (55.3)1577 (49.6)10.4 (3.4)
½–1½29 982 (32.8)1012 (31.9)10.7 (3.4)
2–3½8804 (9.6)429 (13.5)10.8 (3.5)
4+1991 (2.2)154 (4.8)11.1 (3.6)
Missingc97 (0.1)5 (0.2)11.0 (3.5)
Smoking during pregnancy (cigarettes per day)
067 657 (74.0)2352 (74.0)10.6 (3.4)
1–1018 069 (19.8)593 (18.7)10.5 (3.5)
11+5659 (6.2)230 (7.2)10.7 (3.6)
Missingc42 (0.0)2 (0.1)10.9 (3.2)
Coffee consumption during pregnancy (cups per day)
050 477 (55.2)1500 (47.2)10.4 (3.3)
½–7½37 753 (41.3)1468 (46.2)10.7 (3.5)
8+3162 (3.5)206 (6.5)10.9 (3.5)
Missingc35 (0.0)3 (0.1)11.2 (4.0)
Pre-pregnant weight status
Underweight4073 (4.5)164 (5.2)10.6 (3.5)
Normal weight60 982 (66.7)2101 (66.1)10.6 (3.4)
Overweight17 430 (19.1)617 (19.4)10.6 (3.4)
Obese7468 (8.2)263 (8.3)10.5 (3.5)
Missingc1474 (1.6)32 (1.0)10.5 (3.4)
Work schedule during pregnancy
Not working16 778 (18.4)592 (18.6)10.5 (3.6)
Daytime60 141 (65.8)2024 (63.7)10.6 (3.6)
Fixed evening1907 (2.1)72 (2.3)10.5 (3.4)
Fixed night714 (0.8)29 (0.9)10.7 (3.6)
Rotating shift (without night)5958 (6.5)197 (6.2)10.4 (3.4)
Rotating shift (including night)5871 (6.4)255 (8.0)10.4 (3.3)
Missingc58 (0.1)8 (0.3)11.0 (3.7)
Lifting of >20 kg daily during pregnancy
No lifting of >20 kg daily76 458 (83.6)2562 (80.6)10.6 (3.4)
Lifting of >20 kg daily14 170 (15.5)592 (18.6)10.6 (3.5)
Missingc799 (0.9)23 (0.7)11.0 (3.4)
Previously diagnosed genital diseases
No diagnosis of genital disease68 091 (74.5)2258 (71.1)10.6 (3.4)
Ever had cervical conisation1727 (1.9)68 (2.1)10.4 (3.4)
Dysplasia in the cervix1642 (1.8)84 (2.6)10.5 (3.4)
Other genital diseases19 754 (21.6)748 (23.5)10.4 (3.4)
Missingc213 (0.2)19 (0.6)10.4 (3.2)

Figure 2. Number of pregnancies at risk, number of miscarriages, and time of interview at miscarriage according to completed gestational weeks.

Download figure to PowerPoint

Modifiable risk factors

The multiple Cox regression analysis was conducted on 88 373 pregnancies, of which 3079 ended in a miscarriage (Figure 1).

Table 2 shows risk estimates for miscarriage according to potentially modifiable risk factors and confounders. Time-dependent effects on miscarriage were found for seven of the 11 variables in the regression model; risk estimates regarding these risk factors are presented for both the first and second trimester of pregnancy. The majority of risk factors showed the same direction of the association with miscarriage in the two trimesters; however, the associations were more pronounced during the first trimester. We found a hazard ratio (HR) of 3.72 (95% confidence interval [CI], 2.81–4.93) in the first trimester of pregnancy for women aged 40 years or more at conception compared with women aged 25–29 years. In the second trimester, the corresponding HR was just over 2. In both trimesters, dose–response relationships with miscarriage were found for the following risk factors during pregnancy: amount of exercise, alcohol consumption and coffee consumption. Inclusion of previous miscarriage in the model did not change the estimates of the other variables and was not done in the final analysis.

Table 2. Multiple Cox regression analysis including potentially modifiable risk factors for miscarriage and confounders according to the first and second trimester of pregnancy among 88 373 pregnancies in the Danish National Birth Cohort, 1996–2002a
 First trimesterSecond trimester
Pregnancies (= 88 373)Miscarriages (n = 1569)Miscarriages (= 1510)
n (%)bn (%)bAdjusted HR (95% CI)n (%)bAdjusted HR (95% CI)
Potentially modifiable risk factors
Maternal age at conception (years)
<2511 216 (12.7)160 (10.2)1.01 (0.84–1.22)182 (12.1)1.15 (0.96–1.36)
25–2936 587 (41.4)549 (35.0)1.00573 (37.9)1.00
30–3430 170 (34.1)515 (32.8)1.15 (1.02–1.31)556 (36.8)1.16 (1.02–1.31)
35–399463 (10.7)287 (18.3)1.84 (1.58–2.15)168 (11.1)1.07 (0.89–1.29)
40+937 (1.1)58 (3.7)3.72 (2.81–4.93)31 (2.1)2.09 (1.45–3.00)
Amount of exercise during pregnancy (minutes/week)
055 659 (63.0)746 (47.5)1.00809 (53.6)1.00
1–6011 781 (13.3)176 (11.2)1.17 (0.99–1.37)189 (12.5)1.13 (0.96–1.32)
61–1208988 (10.2)225 (14.3)1.83 (1.57–2.13)187 (12.4)1.46 (1.24–1.71)
121–1804932 (5.6)143 (9.1)2.06 (1.72–2.47)127 (8.4)1.79 (1.48–2.16)
181–3004487 (5.1)159 (10.1)2.47 (2.07–2.93)105 (7.0)1.65 (1.35–2.03)
>3002526 (2.9)120 (7.6)3.29 (2.71–3.99)93 (6.2)2.67 (2.15–3.31)
Alcohol consumption during pregnancy (drinks per week)
048 766 (55.2)767 (48.9)1.00758 (50.2)1.00
½–1½29 094 (32.9)469 (29.9)1.05 (0.93–1.18)516 (34.2)1.13 (1.00–1.26)
2–3½8581 (9.7)234 (14.9)1.56 (1.34–1.81)185 (12.3)1.34 (1.13–1.58)
4+1932 (2.2)99 (6.3)2.81 (2.25–3.50)51 (3.4)1.64 (1.23–2.19)
Coffee consumption during pregnancy (cups per day)
048 831 (55.3)718 (45.8)1.00745 (49.3)1.00
½–7½36 510 (41.3)739 (47.1)1.28 (1.14–1.42)684 (45.3)1.22 (1.10–1.37)
8+3032 (3.4)112 (7.1)2.23 (1.79–2.78)81 (5.4)1.84 (1.45–2.34)
Daily lifting of heavy burdens (>20 kg) during pregnancy
No daily lifting of >20 kg74 563 (84.4)1244 (79.3)1.001255 (83.1)1.00
Daily lifting of >20 kg13 810 (15.6)325 (20.7)1.31 (1.15–1.48)255 (16.9)1.03 (0.89–1.18)
Nulliparous41 619 (47.1)680 (43.3)1.00685 (45.4)1.00
Parous46 754 (52.9)889 (56.7)1.15 (1.03–1.29)825 (54.6)1.11 (0.99–1.24)
Highest household occupational status
Higher grade professionals21 144 (23.9)443 (28.2)1.14 (1.00–1.30)356 (23.6)0.91 (0.79–1.04)
Lower grade professionals27 578 (31.2)477 (30.4)1.00499 (33.0)1.00
Skilled workers24 626 (27.9)365 (23.3)0.99 (0.86–1.14)399 (26.4)0.96 (0.84–1.10)
Unskilled workers12 272 (13.9)219 (14.0)1.15 (0.97–1.36)215 (14.2)1.04 (0.88–1.22)
Students2080 (2.4)47 (3.0)1.35 (0.99–1.83)32 (2.1)0.88 (0.61–1.27)
Economically inactive >1 year673 (0.8)18 (1.1)1.67 (1.03–2.71)9 (0.6)0.81 (0.42–1.58)
 Pregnancies (= 88 373)Miscarriages (= 3079)
n (%)bn (%)bAdjusted HR (95% CI)
  1. HR, hazard ratio; CI, confidence interval.

  2. a

    The analysis included both variables with and without time-dependent effect on miscarriage. For the variables with time-dependent effects on miscarriage, the risk estimates and the number of miscarriages are presented for the first and second trimester of the pregnancy, respectively. The risk estimates are adjusted for all variables listed in the table.

  3. b

    Percentages do not add up to 100 in some cases because of rounding.

  4. c

    Estimates are calculated for the entire study period because the risk factor has no time-dependent effect on miscarriage.

Potentially modifiable risk factors
Smoking during pregnancy (cigarettes per day)c
065 616 (74.2)2292 (74.4)1.00
1–1017 362 (19.6)570 (18.5)0.86 (0.78–0.95)
11+5395 (6.1)217 (7.0)0.98 (0.84–1.15)
Pre-pregnant weight statusc
Underweight3995 (4.5)161 (5.2)1.27 (1.08–1.50)
Normal weight59 954 (67.8)2052 (66.6)1.00
Overweight17 097 (19.3)604 (19.6)1.09 (0.99–1.19)
Obese7327 (8.3)262 (8.5)1.15 (1.01–1.31)
Work schedule during pregnancyc
Not working16 249 (18.4)572 (18.6)1.02 (0.92–1.12)
Daytime58 305 (66.0)1970 (64.0)1.00
Fixed evening1801 (2.0)68 (2.2)1.05 (0.83–1.34)
Fixed night670 (0.8)29 (0.9)1.27 (0.89–1.82)
Rotating shift (without night)5713 (6.5)194 (6.3)0.94 (0.81–1.09)
Rotating shift (including night)5635 (6.4)246 (8.0)1.21 (1.06–1.39)
Previously diagnosed genital diseasec
No diagnosis of genital disease 65 984 (74.7)2206 (71.6)1.00
Ever had cervical conisation1674 (1.9)68 (2.2)1.08 (0.84–1.39)
Dysplasia in the cervix 1603 (1.8)80 (2.6)1.43 (1.14–1.79)
Other genital diseases19 112 (21.6)725 (23.5)1.07 (0.98–1.16)

In the Cox regression analysis restricted to pregnancies with information on risk factors collected during pregnancy, the dose–response effect between amount of exercise during pregnancy and miscarriage disappeared, and no effect of smoking on miscarriage was found. Compared with the main analysis, an opposing tendency was found in the sub-cohort for dysplasia in the cervix, as this diagnosis had a protective effect on miscarriage risk (data not shown).

The analysis restricted to primigravid women with a short time to pregnancy showed some deviations compared with the main results; the effect of alcohol consumption during pregnancy on miscarriage was even more pronounced, but the increased risk of miscarriage associated with obesity as well as the protective effect of smoking disappeared. An increased but statistically insignificant effect on miscarriage was found for cervical conisation and a slightly increased risk was associated with dysplasia in the cervix, respectively, compared with no history of genital diseases (data not shown).


Table 3 shows the results of the PAF analyses. In a prevention scenario where the women conceived at an age of 25–29 years, consumed no alcohol during pregnancy, were normal weight before pregnancy, did not lift >20 kg daily during pregnancy, and worked only during the day, 25.2% of the miscarriages were preventable. This scenario resulted in the highest preventable proportion of miscarriage. Among the single risk factors, the highest proportion of miscarriages were preventable if all women conceived when aged 25–29. Alcohol consumption was the most important risk factor during the pregnancy. Inclusion of induced abortion as an event together with miscarriage in the analysis did not change the estimations of PAFs (data not shown) considerably.

Table 3. Preventable proportions of miscarriages according to different prevention scenarios defined by single and multiple potentially modifiable risk factors among 88 373 pregnancies in the Danish National Birth Cohort, 1999–2002a
Prevention scenariosPAF %b
  1. a

    The population attributable fraction (PAF) estimates are based on the assumption that the defined prevention scenario is fulfilled for all 88 373 pregnancies under study.

  2. b

    The PAFs are not additive, as the women can have more than one risk factor.

Defined by single potentially modifiable risk factors before pregnancy
Maternal age of 25–29 years at conception11.4
Pre-pregnant normal weight3.7
Defined by single potentially modifiable risk factors during pregnancy
No consumption of alcohol9.0
No lifting of >20 kg daily2.4
Daytime work1.5
Defined by multiple potentially modifiable risk factors before pregnancy
Maternal age of 25–29 years at conception and pre-pregnant normal weight14.7
Defined by multiple potentially modifiable risk factors during pregnancy
No consumption of alcohol, no lifting of >20 kg daily, and daytime work12.5
Defined by multiple potentially modifiable risk factors before and during pregnancy
Maternal age of 25–29 years at conception, no consumption of alcohol during pregnancy, pre-pregnant normal weight, no lifting of >20 kg daily during pregnancy, and daytime work during pregnancy25.2


  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Conclusions
  8. References

Main findings

In this large nationwide follow-up study, we found that 25.2% of the miscarriages were preventable by modification of multiple risk factors to low risk levels. Maternal age at conception and alcohol consumption during pregnancy were the single most important risk factors. Our study extends the relative risk approach used in most previous studies by estimating preventable proportions of miscarriage attributed to multiple modifiable risk factors.


The increased risk of miscarriage with advanced maternal age was consistent with previous studies.[3] We found alcohol consumption during pregnancy to be an important risk factor. Alcohol has been identified as a risk factor for miscarriage in previous studies,[5-7] but in contrast to our findings some studies found an effect only in the first or second trimester.[5, 7] The amount of alcohol during pregnancy that is harmful to the fetus also varies between studies.[5-7] In contrast to our findings, it was concluded in a review that alcohol abstinence was unlikely to prevent miscarriage.[2] Although the association between alcohol and miscarriage in our study might be explained by reverse causation,[4, 25] it could be that alcohol and its metabolites lead to fetal chromosomal abnormalities that result in miscarriage.[26]

Contrary to the present study, smoking was found to explain relatively high proportions of miscarriage in some previous studies.[12, 13, 15] In a more recent cohort study, however, no association between smoking and miscarriage was identified.[10] We believe that our finding of a beneficial effect of smoking one to 10 cigarettes per day on miscarriage is a result of chance, as smoking of more than 10 cigarettes per day was not associated with increased risk. However, it could be explained by behaviour modification bias if women with previous miscarriages have stopped smoking to increase their chances of successful childbearing.

Our results confirm previous findings of pre-pregnant underweight[14, 27] and obesity[28] as risk factors for miscarriage. However, some studies were not able to find an association between weight status and miscarriage.[29] In our study, obesity was not associated with increased risk of miscarriage among primigravid women with short time to pregnancy. An explanation can be that obese women are less fertile and this condition in itself, or fertility treatment, is a known risk factor for miscarriage.[22]

In line with our results, previous studies have demonstrated that heavy lifting and night work increase the risk of miscarriage.[30, 31] These findings suggest that it may be relevant to introduce specific rules regarding the employers' responsibility to protect the pregnant women from the elevated risk of miscarriage.

The effect of amount of exercise on miscarriage disappeared when we restricted the analysis to women who were interviewed during the pregnancy. If the women who were interviewed after a miscarriage had systematically better recall regarding exercise compared with women with other pregnancy outcomes, this may have created a spurious association with miscarriage. In the DNBC, a positive association has been identified between exercise and miscarriage in gestational week 11–14 in a sub-cohort of women interviewed during pregnancy.[32] It suggests that the effect of exercise is restricted to the earliest period of pregnancy. However, another study found no association between physical exercise during the first trimester and miscarriage.[14]

In our study, coffee consumption was associated with miscarriage, which is consistent with findings in a recent cohort study that adjusted for nausea.[9] In a meta-analysis, presence of pregnancy symptoms was related to a 64% decreased risk of miscarriage compared with absence of symptoms.[33] In a case–control study, adjustment for nausea removed the entire association between coffee consumption and miscarriage.[14] We had no information on nausea, and reverse causation could explain the dose-response relation found in the current study.

Contrary to our expectations, cervical conisation was not a risk factor for miscarriage.[34] Instead, we found an association between miscarriage and dysplasia in the cervix, but any biological mechanism behind this association is difficult to understand.

Strengths and limitations

Our study has considerable strengths. We were able to provide estimates with substantial power because of the large study sample and a high number of miscarriages. Early recruitment of pregnancies into the study made it possible to present risk estimates from six gestational weeks. As detailed self-reported information on modifiable risk factors and confounders was available, it was possible to use a multiple risk factor approach and compare the effect of different risk factors on miscarriage prevention. The Danish registers enabled optimal linkage of information and almost complete follow-up. Lastly, we used survival analysis methods, allowing inclusion of pregnancies at different gestational ages and the use of time-dependent exposure measures.

Our study also has some limitations. Although we have early recruitment to the DNBC, the majority of women were recruited after the gestational age where miscarriage is most common; this explains why the observed proportion of miscarriages was only 3.5%, in contrast to the 11–14% of all pregnancies which are expected to end in miscarriages.[35] A fetal life-table analysis on the cumulative risk of fetal death has been estimated from week 6 and was found to be 11%[36] and the use of survival analyses with left truncation enables bias related to the variation in recruitment to be taken into account. Compared with other pregnancy outcomes, an increased proportion of the DNBC participants whose pregnancy ended in an early miscarriage decided not to participate. The outcome probably influenced some women's participation, resulting in biased estimates, but it is not known whether the exposures influenced participation. As excluding subjects from survival analysis on the basis of future events would bias results, we allowed women whose pregnancy ended in an induced abortion to contribute with risk time until event. Censoring due to induced abortion was likely to represent fetuses with systematic reduced changes of survival. However, the sensitivity analysis indicated that the handling of induced abortions did not seem to affect the conclusions. Another limitation was that the majority of the women who experienced a miscarriage were interviewed after the event, which could have introduced recall bias. The validity of the PAF estimates depends on some methodological assumptions, i.e., causality, non-confounding and well-defined interventions, and these requirements were not necessarily fulfilled for all risk factors in our analysis. The associations between amount of exercise during pregnancy, coffee consumption during pregnancy, previously diagnosed genital diseases and miscarriage were likely to be biased because of systematic differences in recall of exposure and residual confounding; consequently, we defined no prevention scenarios regarding these factors. However, even for the risk factors included in the final analysis we cannot tell from this observational study whether the associations are causal. This is partly because confounding or reverse causality could have resulted in biased risk estimates, except from those presented for maternal age. How easy it is to modify maternal age at time of conception may be questioned. Although factors such as having a partner, education, social network, and the norms in the Danish society play an important role to the modifiable aspects of this risk factor, we think that information about increased chances of having a successful pregnancy in a relatively young age as well as the society's facilitation of parenthood at an early age by creating better opportunities for combining education or participation in the working force and childbearing are important perspectives in future miscarriage research. The women in the DNBC are not a representative sample of all Danish pregnant women.[37] Trends of lifestyle-related risk factors have changed since the DNBC was established but were not taken into account when estimating the preventable proportions of miscarriages. This could have influenced the generalisability of the results.


  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Conclusions
  8. References

This study adds an important perspective to the miscarriage research by suggesting that some miscarriages are preventable. A future study based on similar methods but using relative risk estimates obtained from meta-analysis would be a preferred approach to estimate PAFs. If our findings are supported by future prospective cohort studies, they may be used in a prevention strategy against miscarriage.

Disclosure of interest


Contribution to authorship

AMNA, PKA and SFN were responsible for the concept of the study. AMNA obtained the funding and acquired the data. SFN did the analyses, interpreted the results and wrote the first draft of the manuscript. PKA, KSL and AMNA supervised the analyses and revised the manuscript. All the authors discussed and approved the final manuscript.

Details of ethics approval

Ethical approval is not required for registry-based studies in Denmark.


This work was supported by the University of Copenhagen (grant number 3306045). The Danish National Research Foundation has established the Danish Epidemiology Science Centre that initiated and created the Danish National Birth Cohort. The cohort is furthermore a result of a major grant from this foundation. Additional support for the Danish National Birth Cohort is obtained from the Pharmacy Foundation, the Egmont Foundation, the March of Dimes Birth Defects Foundation, and the Augustinus Foundation.


We are particularly grateful to all of the women who participated in the Danish National Birth Cohort.


  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Conclusions
  8. References

Causation or association: running before we can walk?

Causation or association: running before we can walk?
  • K Hemming

  • Public Health, Epidemiology and Biostatistics, University of Birmingham, Birmingham, UK

Linked article: This article is a mini commentary on Feodor Nilsson S et al., p. 1375–84 in this issue. To view this article visit

Nilsson et al. estimate that about 25% of miscarriages might be avoidable (BJOG 2014; DOI: 10.1111/1471-0528.12694). These estimates come from the analysis of a large and well executed cohort study. The analysis includes outcomes of more than 90 000 pregnancies, of which 3177 resulted in miscarriage. Estimating the population attributable risk, they determine that about 25% of miscarriages might be avoidable if all women conceive between the ages of 25 and 29 years; abstain from alcohol; do not lift heavy weights or work night shifts. This finding has important public health implications, as for example in the UK each year there are approximately 50 000 admissions for early pregnancy loss (

However, in this study information on possible risk exposures (i.e., alcohol intake, smoking, etc.) was elicited after the miscarriage had occurred in 2444 of the women. Nilsson and colleagues did carry out a sensitivity analysis, excluding cases in which the exposure data were collected after the miscarriage; some differences were observed, although numerical results are not reported. Elicitation of risk factors after the event (miscarriage) has occurred, might be subject to recall bias. For example, women who have had a miscarriage might be more inclined to attribute their miscarriage to exposures and so more likely to report exposures that those who had not experienced a miscarriage might deem inconsequential. Unfortunately, the direction of this bias is not predictable and might lead to either attenuation of effects or spurious associations (Straus et al. Evidence-based Medicine: How to Practice and Teach It, 4th edn. London: Churchill Livingstone, 2010).

Reporting attributable risk allows the magnitude of effects to be estimated in absolute terms and so for this reason might be preferred over the reporting of relative risks. However, the term ‘attributable’ implies causation. Cohort studies are limited in their ability to infer causation because of the possibility of residual confounding, and so can only robustly provide information about associations. Some casual associations have been established from cohort studies, but evidence from other sources is needed before the leap to causation can be made. Indeed, many associations identified in cohort studies have later been shown not to be causal associations. The classic example is the use of hormone replacement therapy and risk of cardio-vascular events, where earlier protective associations observed in observational studies were later identified to be due to confounding (N Krieger et al. J Epidemiol Community Health 2005;59:740–8).

In this study, arguably the least biased estimates would come from an analysis restricted to those for whom elicitation of exposures was done before the miscarriage. However, this would have been at the expense of excluding half of the events and so would have had implications for power. Good practice reporting guidelines exist for observational studies (Strobe). The equivalent reporting guidelines for randomised control trials (CONSORT) give a clear preference for an intention-to-treat analysis over a per protocol analysis, as the latter can induce biases. Observational studies are very varied in nature and so very prescriptive reporting guidelines are not possible. But reporting of results which are probably least biased, while not part of Strobe, might be considered good practice in the reporting of observational studies.

Disclosure of interests

I am a paid statistical reviewer for BJOG; other than that, I have no competing interests to declare.

  • K Hemming

  • Public Health, Epidemiology and Biostatistics, University of Birmingham,

  • Birmingham, UK