Late miscarriage and preterm birth after treatment with clindamycin: a randomised consent design study according to Zelen


Dr P-G Larsson, Department of Obstetrics and Gynaecology, Kärnsjukhuset, 541 47 Skövde, Sweden. Email


Objective  To screen for bacterial vaginosis (BV) and to investigate the effect of treatment with vaginal clindamycin in order to observe the effect on late miscarriage and delivery prior to 37 completed weeks (primary outcome).

Design  Randomised consent design for clinical trials according to Zelen.

Setting  Southeast region of Sweden.

Population  A total of 9025 women were screened in early pregnancy.

Methods  A total of 819 women with a Nugent score of 6 and above were considered to have BV and treated according to Zelen allocation. The incidence of late miscarriage and spontaneous (noniatrogenic) preterm birth was assessed.

Main outcome measures  Late miscarriage and spontaneous preterm delivery before 37 weeks.

Results  Therapy with vaginal clindamycin had no significant impact on the incidence of spontaneous preterm delivery prior to 37 completed weeks; OR 0.90, 95% CI 0.40–2.02 (primary outcome variable). However, only 1 of 11 women in the treatment group versus 5 of 12 in the control group delivered prior to 33 completed weeks; OR 0.14, 95% CI 0.02–0.95. Treatment was associated with 32 days longer gestation for the 23 participants who had late miscarriage or spontaneous preterm birth (P= 0.024, Mann–Whitney U test) and significantly fewer infants had a birthweight below 2500 g (secondary outcome). A follow up of infants born preterm 4 years postnatally indicated that extending gestational age did not increase the number of sequelae.

Conclusions  Clindamycin vaginal cream therapy was associated with significantly prolonged gestation and reduced cost of neonatal care in women with BV. Early screening for BV and treatment with clindamycin saved approximately €27 per woman.


Preterm birth prior to 37 completed weeks is an important cause of perinatal mortality and is responsible for about half of all perinatal morbidity. The earlier the birth, the higher the risk of mortality or permanent injury. In Sweden, three-quarters of preterm deliveries are spontaneous as opposed to induced or iatrogenic preterm birth.1 A significant cause of both preterm birth and late miscarriage is infection in the amniotic fluid, which has its highest incidence between 22 and 28 gestational weeks, declining thereafter.2 For nearly 15 years, the correlation between bacterial vaginosis (BV) and incidence of late miscarriage and spontaneous preterm birth has been controversial. One meta-analysis3 indicated that BV is associated with a two-fold increase in the risk of spontaneous preterm birth; OR 2.19, 95% CI 1.54–3.12. Many studies have explored a variety of treatments to reduce the risk of spontaneous preterm birth associated with BV.4,5 There have been 13 studies published on treatment of BV and 13 meta-analysis. A Cochrane Systematic Review6 failed to show any benefit from treatment, but three subsequent treatment studies reported that it reduces the incidence of spontaneous preterm birth.7–9 However, in modern practice, delivery prior to 37 completed weeks but at or after 33 completed weeks has a low morbidity and mortality. The risk of neonatal problems is mainly associated with birth before 33 completed weeks. In the UK, 90% of these children are transferred to special care baby units as opposed to only 14% born at 37 completed weeks.10 In Sweden, 71% of the preterm infants are born at 35–36 weeks; including these gestations in outcome measures might obscure important beneficial effects of early antibiotic treatment. In Sweden, the frequency of delivery prior to 33 completed weeks is only about 1%.1,11

The primary objective of our study was to screen for BV and treat asymptomatic women according to Zelen randomisation and to observe the effect on delivery prior to 37 completed weeks (primary outcome). However, given that the major perinatal risk is associated with delivery prior to 33 completed weeks, the study also analysed the effect of treatment on the incidence and associated morbidity of such births (secondary outcome).

Materials and methods

The study was conducted in the Southeast Health Care Region of Sweden. Participants recruited were between 10 and 14 weeks of gestational age at their initial visit to the antenatal clinic. If the woman agreed to participate in the study, a specimen of vaginal fluid was collected by the physician or the midwife from the posterior or the lateral vaginal wall and placed on a microscopic slide. The slides were air-dried and analysed in the Department of Clinical Microbiology at Linköping University Hospital, using Gram staining. Examination was performed by an experienced microscopist (B.C.). The specimen slides were scored according to Nugent criteria using at least five microscopic fields of less than ×1000 magnification.12

Halfway through the study, a change over to a microscope with a 40% larger image area made it necessary to re-evaluate all Nugent-scored intermediate slides and conduct a complementary evaluation according to the Hay/Ison criteria.13,14 Re-evaluation was performed by two of the investigators (B.C. and P.-G.L.). The size of the microscope image area has a bearing on Nugent scoring,15 and women in the intermediate flora group with a Nugent score of 6 were assigned to the BV group on the assumption that a Nugent score of 6 indicates a true group of women with BV as reported previously.15 In the classification according to Hay/Ison, the image field size does not directly affect the score because the evaluation is based on a ratio rather than the exact number of bacteria. Therefore, we used Hay/Ison criteria to verify our diagnoses. Slides of the smears from women infected with BV with more than one Lactobacillus morphotype per microscopic field were scored as intermediate because this gave two points for the Lactobacillus morphotype and a maximum of four points for more than 30 Gardnerella morphotype per field (although the field contains more than 10 000 000 bacteria of Gardnerella morphotype and possibly even clue cells), thus a score of 6.8,10 A microscope with a larger image area increases the number of bacteria viewed per microscopic field. This is not significant in normal smears with more than 30 Lactobacillus morphotypes but has a considerable impact if the counts of Lactobacillus morphotype per microscopic field increase from less than 1 to 1–4, which it can easily do with a bigger image area. The original evaluation was therefore used except that women with a Nugent score of 6 were transferred to the BV group. At the time of re-evaluation, the pregnancy outcome data were still unknown.

Enrolment of participants to the study began in January 1999 and continued until May 2001. During this period, approximately 20 000 women were registered in the antenatal clinics, and of these 9025 consented to participate in the study. All women with BV were randomised to either an intervention group with a 7-day regimen of treatment with clindamycin vaginal cream or a control group to remain untreated and uninformed of their BV status as stipulated in the prerandomised consent design for clinical trials. Only women who were diagnosed with BV and randomised to the intervention group were told the result of their vaginal smear. Clindamycin treatment usually started within a week of diagnosis. Clinicians caring for the BV-positive group randomised to nontreatment were not told of the diagnosis.

Women in the intervention group were re-examined twice, in gestational weeks 24 and 31 (±2 weeks), and a new vaginal specimen was taken, air-dried and analysed. If the examination showed the presence of intermediate flora or BV, the women allocated to the treatment group were prescribed another 7-day course of clindamycin cream.

Pregnancy data were collected through the Medical Birth Register (MBR) at the Swedish National Board of Health and Welfare. Any missing data were manually extracted from individual medical records through the personal identification number (PIN) assigned to all legal residents at birth or after immigration. The PIN is used to establish links between different registers and makes it possible to trace women who had changed their residence and given birth in some other part of the country.

In the southeast region of Sweden where this study was conducted, all pregnant women attend antenatal clinics that are part of the public healthcare system. There is no private antenatal care alternative. Therefore, all women studied had similar healthcare management throughout pregnancy. During the study period, 22 997 pregnant women gave birth in our region and about half of these participated in the study. However, not all antenatal clinics started recruiting women at the same time, and some antenatal clinics were closed during the summer months, which means that some women paid an initial antenatal clinic visit when the pregnancy was too advanced for their inclusion in our study. A consequence of this is that overall enrolment to the study was lower during the summer months.

Classification into spontaneous preterm birth and induced or iatrogenic preterm birth followed the diagnosis registered in the MBR according to ICD-10, which was performed with no foreknowledge of the scoring of the Gram-stained vaginal smears (although the birth attendants would have known if a course of clindamycin had been given). Iatrogenic preterm birth was defined as either induction of labour or elective or emergency caesarean section prior to 37 completed weeks and was usually due to a diagnosis such as moderate or severe pre-eclampsia (O140–9), placental abruption (O45), placenta praevia (O44) or other maternal complications such as pelvic girdle syndrome or intrahepatic cholestasis (O26) and fetal complications such as fetal growth restriction (O365). In the MBR, these diagnoses are not hierarchically classified. To obtain a hierarchical ranking, we used Microsoft Access software to establish the most important diagnosis. Any woman with preterm prelabour rupture of membranes was classified as having had a spontaneous preterm birth even if labour was induced after 37 completed weeks. Expected delivery dates were based on best estimate from the MBR. This in turn is based on ultrasound scanning in the second trimester at 16 or 18 gestational weeks and if this data are missing, on the date of the last menstrual period.1

We defined early miscarriage as spontaneous pregnancy loss prior to 16 completed gestational weeks. The definition of missed miscarriage is discovery through ultrasound scanning at 16–18 gestational weeks of no fetus or a nonviable fetus of a size equivalent to a mean for gestational age of less than 16 completed weeks. Late miscarriage was defined as spontaneous miscarriage after 16 weeks but before 22 completed weeks.

The criteria for inclusion in the study were the following: women aged 18 or older, Swedish speaking, no antibiotic treatment in early pregnancy and no symptomatic vaginal infection.

The criteria for exclusion from the study were the following: therapeutic termination of pregnancy, early spontaneous miscarriage or missed miscarriage, postinclusion need for cervical cerclage or postinclusion treatment with either metronidazole or clindamycin outside the study. Women with multiple pregnancies (twins or triplets) were also excluded. From the MBR, data regarding treatments with other antibiotics during the pregnancy were extracted to make sure that none had received treatments elsewhere.

The cost-benefit analysis gave an estimated cost of €1000 for 1 day of intensive care at the neonatal ward. This does not include the cost of managing sequelae. All preterm infants were followed up at routine health screening at the age of 4.

Study statistics

This was a randomised consent design clinical trial according to Zelen.16,17 Randomisation was performed by the Department of Clinical Microbiology at the University Hospital in Linköping after evaluation of the vaginal smear and was computer generated in blocks of ten. Some of the women who were randomised to the intervention group declined medication but were (according to Zelen) allocated to the treatment group and analysed according to intention to treat. The statistical analyses used were survival analysis according to Kaplan–Meier, Mann–Whitney U tests, chi-square test, odds ratio and Fisher’s exact t test (SPSS for Windows version 12.0; SPSS Inc., Chicago, IL, USA). The level of statistical significance was set to 0.05.

The statistical power of our study was calculated on the assumption that there is a two-fold increased risk of preterm birth among women with BV. As the expected incidence of spontaneous preterm birth is 8% among women with BV and treatment was assumed to lower this to 4%, 600 women were needed in the intervention group to achieve a power of 80% to detect a difference significant at 5%. On the assumption that the prevalence of BV would be 15%, it was necessary to screen 10 000 women.


The randomised consent design clinical trial according to Zelen was approved by the regional Ethics Committee in Linköping and the Swedish Medical Products Agency, with the modification that all women were informed about the study procedure during the initial visit to the antenatal clinic. In a separate application, the follow up of the preterm infants was approved by the Regional Ethics Committee in Linköping.


A total of 9025 women were screened for BV. Three women who were initially included in the study were withdrawn because of cervical cerclage. Another woman was withdrawn as she had undergone, prior to the screening, treatment with clindamycin vaginal cream because of symptomatic BV. Of the remaining 9021 randomised women, 13 were lost to follow up. One, who was a refugee, could not be traced; six had left Sweden; three proved not to be pregnant, one suffered fatal cardiac failure and two chose to be delivered in at home. A further 217 were excluded for the following reasons: in two cases, Gram-stained smears were lost or were not satisfactory for analysis; 40 underwent therapeutic termination of pregnancy; 145 had an early spontaneous miscarriage and 30 had a missed miscarriage (Figure 1).

Figure 1.

Flow chart of all women in the preterm delivery study. At the start, there were 9025 women, and 8791 remained after exclusion.

Following Gram staining, 6904 smears were classified as normal vaginal flora, 1178 as intermediate flora and 709 as showing BV according to Nugent. Of the 1178 slides with intermediate flora, 110 with a Nugent score of 6 were subsequently reallocated to the BV group.

Among the 819 women with BV (Nugent score ≥6), 55 who were randomised to the intervention group declined medication but were analysed according to intention to treat; thus, 408 women were classified as treated and 411 as nontreated. The mean age of the women in the intervention group was 28.5 years (SD 4.83), and the mean age of the women in the control group was 28.6 years (SD 4.97). Both the intervention group and the control group included 45.5% nulliparae. The incidence of previous preterm birth among the parous women in the intervention group was 9.2% (20/217) as compared with 6.0% (13/218) among those in the control group (P= 0.20, chi-square test). The intervention group included 76.2% of nonsmokers compared with 77% in the control group. However, there were more heavy smokers (≥ ten cigarettes per day) in the intervention group, 10.2 versus 6.6% (P= 0.073, chi-square test).

Preterm birth

There were 46 cases of delivery prior to 37 completed gestational weeks, 21 (5.1%) in the intervention group and 25 (6.1%) in the control group. The odds ratio for preterm birth in the intervention group was 0.84 (95% CI 0.48–1.47).

Spontaneous preterm birth

After excluding 19 multiple pregnancies, 3 women who were treated outside the study with metronidazole or clindamycin and 12 iatrogenic or induced preterm deliveries, there remained 11/395 (2.8%) late miscarriages or spontaneous preterm deliveries in the intervention group versus 12/390 (3.1%) in the control group. The odds ratio for late miscarriage or spontaneous preterm birth prior to 37 completed weeks and after treatment with clindamycin was 0.90 (95% CI 0.40–2.02).

The 11 cases of late miscarriage or spontaneous preterm birth in the intervention group had a mean gestational length of 247.6 days at delivery compared with 215 days among the 12 cases in the control group, a difference of 32.5 days (Mann–Whitney U test, P= 0.024) (to achieve this, prolongation of pregnancy required 408 women to be treated). Kaplan–Meier survival curves for women with late miscarriage or spontaneous preterm birth prior to 37 completed weeks showed a significant prolongation of the gestational length, with a log rank of 4.24 (P= 0.0396) (Figure 2).

Figure 2.

Survival functions (Kaplan–Meier) of gestational days at birth for women with late miscarriage and spontaneous preterm birth prior to 37 completed weeks (<259 days).

In the intervention group, only 1 of 11 spontaneous preterm deliveries occurred prior to 33 completed weeks as opposed to 5 of 12 in the control group (OR 0.14 [95% CI 0.02–0.95]).

Of the 21 live births resulting from spontaneous preterm delivery prior to 37 completed weeks, 12 infants did not require treatment other than normal neonatal care. Of the nine infants treated in the neonatal intensive care unit, those whose mothers were in the control group required a mean number of 45 days of care (range 14–94 days, n= 5), while those whose mothers were in the intervention group required a mean of 18 days of care (range 4–39 days, n= 4) (Mann–Whitney U test, P= 0.14). The cumulative days of neonatal intensive care for the infants in the control group was 223 versus 70 for those in the intervention group, a difference of 153 days (Kaplan–Meier survival analysis, log rank 2.14 (P= 0.14, Figure 3). With an estimated cost of €1000 per day at the neonatal care unit, these 153 days entailed an additional cost of €153 000 for the nontreated women, i.e. the control group. The mean birthweight of the ten infants born in the control group was 2177 versus 2635 g for the 11 infants in the intervention group, a difference of 458 g (Mann–Whitney U test, P= 0.067). Significantly, fewer infants with birthweight less than 2500 g were found in the intervention group (P= 0.009, Fisher’s exact test).

Figure 3.

Number of days in the neonatal intensive care unit for nine premature infants born through spontaneous preterm delivery (<259 days) by mothers with BV (Nugent score 6–10).

The 4-year routine examination of the 21 premature infants demonstrated very few sequelae. Among the children in the control group, one was diagnosed with retrolental fibroplasia with a need for strong glasses and one was diagnosed with bronchopulmonary dysplasia in the neonatal period but is now considered well (Table 1). No severe treatment-related adverse events was noted, but three minor adverse events were recorded as three women withdrew from treatment because of persistent itching.

Table 1.  Births before 37 completed weeks of gestation in women with BV
TreatmentDelivery daysWeightHospital daysAdmission to neonatal intensive care unit4-year follow up
No treatment119 
17581061YesBronchopulmonary dysplasia
188102094YesRetrolental fibroplasia
235314017YesBeckwith-Wiedeman’s syndrome


This study did not show any significant reduction in the incidence of late miscarriage and spontaneous preterm birth prior to 37 completed weeks (OR 0.90, 95% CI 0.40–2.02) (primary outcome variable) associated with the treatment of BV with vaginal administration of clindamycin cream. However, treatment was associated with a 32 days longer gestation in babies born preterm (Mann–Whitney U test P= 0.024) (secondary outcome variable). This probably has greater clinical significance than simply the incidence of preterm birth. The study also found that the number of infants with a birthweight less than 2500 g was significantly less in the intervention group. These results are in accordance with previous studies.7,9

Care of premature infants in a neonatal intensive care unit is complicated and expensive. Our study suggests that clindamycin treatment is associated with cost benefits. The cost of screening and treating BV is far less than the additional cost incurred by care in the neonatal intensive unit. Moreover, the lifetime cost of premature infants is higher still due to the increased risk of late neurodevelopment morbidity.18 Two infants born to nontreated mothers in the control group had late sequelae, albeit of a minor degree. An important observation in our study is that there was no suggestion that prolongation of gestation among clindamycin-treated mothers increased the risk of adverse sequelae, although our population is too small for any firm conclusion in this respect.

Preterm birth accounts for 50% of newborn morbidity. However, morbidity differs enormously depending on whether the infant is born prior to 33 completed weeks or between 33 and 37 completed weeks. Finnström et al.19 showed that between 22 and 26 weeks of gestation, everyday that the baby remained in utero increased survival by 3%. Delivery at 33–37 completed weeks is therefore a less important outcome measure than birth earlier than this. Delivery prior to 33 completed weeks carries a much higher risk of complications; thus, classification of the specific gestational week in which a preterm infant is born is more important than simply assessing the proportion of births prior to 37 completed weeks. This fact might account for the contradictory results presented in previous studies of antibiotic treatment intended to reduce preterm birth among women infected with BV.4–6,20 Most preterm infants in Sweden are born in the interval 34–36 completed gestational weeks, which might bias the statistics and obscure the positive effects of early antibiotic treatment to prolong pregnancy. National figures in Sweden show that there has been a recent decline in singleton preterm births in the interval 34–36 completed gestational weeks nearest term but not for births <34 weeks.1

As only half the potential recruits were included in our study, we did not reach our planned recruitment; thus, our study is relatively underpowered. However, we have no indication that this caused any significant bias in our results, as during the time of the study, the overall incidence of preterm singleton birth in our region was 5.56% (n= 22 291) compared with 5.48% among the 8547 singletons in our study population.

Treatment of asymptomatic BV remains controversial. Early spontaneous preterm birth is more likely to be of infectious aetiology than late preterm birth.21 In recent reviews, the assumption is generally that antibiotics for the treatment of BV should be used early in gestation, before its effects become irreversible.20,21 In our study, we initiated treatment at a mean gestational age of 96 days (weeks 13 + 6, SD 18 days), and we recruited 90% prior to 16 gestational weeks (some women were treated late because they were assigned according to the date of their last menstrual period and before their gestation was corrected by ultrasound scan). In recently published studies, treatment started at mean and maximum gestational ages at screening of 16–17 weeks and 19–22 weeks, respectively,7–9 and this might account for differences in the reported results.

Resolution of BV after one treatment at 23 completed weeks was 70%; thus, 30% received a second treatment. At the second follow up at 31 completed weeks, 10% had relapsed and were again treated with clindamycin. These figures are in accordance with those of Lamont et al.7 The first course of treatment in their study was only for 3 days, but during follow up, any repeat treatment was given using a 7-day course. Our results, with treatment initiated early and repeated when abnormal flora was detected at follow up, was associated with a significant prolongation of pregnancy and a reduction in very preterm birth and late miscarriage. Neither our study nor the Lamont or Kiss studie supports the idea that it is necessary to administer clindamycin orally as suggested by Ugwumadu et al.8

Our study has both advantages and limitations. Although our population was relatively large, only 23 of the women with BV had a spontaneous preterm delivery; however, this reflects the fact that the incidence of preterm deliveries in Sweden is among the lowest in the world. Another factor which contributed to the small number of spontaneous preterm deliveries was that only 9% of the women in our study had evidence of BV at the first antenatal visit.

Despite the relatively small number of spontaneous preterm births, we were able to demonstrate a potential benefit of treatment with clindamycin as gestation was prolonged by 32 days. The number needed to treat to achieve one prolonged gestation was 37. Our estimation of the power of our study was based on an estimate of the prevalence of BV (15%) that was too high. We also based our calculation on an estimate of preterm births in the nontreated women that was also too high. Our calculations were based on data from the late 1990s and were the best data then available. From our more recent meta-analysis, we suggested that treatment of BV in early pregnancy could significantly reduce spontaneous preterm birth, with an odds ratio of 0.83 (95% CI 0.71–0.96).5 This analysis suggested that a prospective trial should include 9506 women with BV, of which half should be treated. Results that are in accordance with the recently published meta-analysis with a RR of 0.73; 95% CI 0.55–0.98) and based on 6188 women with BV.22 Our study with only 819 women is thus far too small to show, by itself, a conclusive result.

Our cost-benefit analysis showed that the cost for the extra 153 days in the neonatal intensive care unit for the untreated children was €153 000. Thus, for a population of approximately 9000 pregnant women, the additional cost of neonatal care comparing no screening or treatment with universal screening and treatment would be €306 000.

This cost can be viewed in relation to the cost of screening and treating 9000 women for BV, where the individual cost is around €3.30 (total cost €29 700) if the diagnosis is according to clinical criteria. If the diagnosis is performed by collecting Gram-stained smears, which are analysed at a laboratory according to Nugent criteria, the individual cost will be €9 (or for 9000 cases, €81 000). Clindamycin treatment costs €18 per women and treatment of 819 cases will consequently be €14 742. In sum, screening and clindamycin treatment will cost €44–€95 000 for our 9000 women. This would save the public healthcare system €23 per women according to our study or a total of at least €211 000. If this is put into general practice in Sweden, public health savings will be €2 300 000 per year, discounting any further cost for sequelae related to premature delivery.

We modified the Nugent score for diagnosis of BV as we have discussed in a previous study5 since we encountered technical limitations (influence of image area on scores) in the Nugent scores. For this study, we therefore validated our modification of the Nugent score by comparing it with the Hay/Ison classification, showing a kappa index of 0.884.

The study by Ugwumadu et al.8 found an increase of pre-eclampsia in the intervention group, i.e. the group treated with clindamycin. Our study found five gravid women with moderate or severe pre-eclampsia that resulted in preterm birth. Four of these were in fact randomised to treatment, although two declined; thus, we have no indication that treatment with clindamycin led to pre-eclampsia.

Every legal resident of Sweden is assigned a PIN, which is used in many contexts including health care. The PIN is used to establish links between different data registers and makes it possible to trace an individual and legally to obtain relevant information on mothers and children. The information which was missing in the MBR could be extracted from the individual medical records. Of the women included in the study, all but one could be traced, although six had moved abroad and thus were lost. Using the PIN, it was possible to trace all children as well. Our study is the first to investigate the outcome for neonates born to mothers who during pregnancy had received treatment for BV, with the objective of reducing preterm birth. We were able to show that prolongation of the gestation had no significant negative effect on the child’s health.


This study has received grants from the Medical Research Council of Southeast Sweden and Linköping University. We wish to thank members of the premature study group of Southeast Sweden and all midwives at the antenatal clinics for enrolling women in the study; thanks also to the staff at the MBR, for statistical support. Our appreciation goes to Pharmacia Ltd, Uppsala, for supplying our study with clindamycin vaginal cream. All members of the premature study group of Southeast Sweden work at a Department of Obstetrics and Gynaecology: Christina Gunnervik, Värnamo; Wivi-Anne Holgersson, Västervik; Ann Josefsson, Linköping; Lisbeth Liest, Eksjö-Nässjö; Ditte Pehrsson-Lindell, Motala; Mona-Lisa Polland, Jönköping; Gunnar Thorbert, Kalmar; Agneta Werner, Norrköping.

Conflict of interest

P.-G.L. has received payment for lectures and funding for trials from Pharmacia Ltd.

Contribution to authorship

P.-G.L., L.F., T.J., B.C. and U.F. contributed to design and collection of data, P.-G.L., L.F. and U.F. contributed to interpretation of results, statistical analysis and writing of the report.