Mums thinking about homebirth were greeted this week with a host of conflicting headlines regarding its safety in the UK.

Some were alarming.

Home birth ‘carries higher risk’ for first-time mothers – The BBC

Home birth risks up for new mums – The Sun

First time mothers warned over home birth risks – The Telegraph

First-time mothers who opt for home birth face triple the risk of death or brain damage in child – The Daily Mail

Others, not so much.

Study finds home birth is safe – NHS News

Home births: ‘Women need help to make the right choice’ – The BBC

Home as safe as hospital for second births – The Independent

Women with low-risk pregnancies ‘should have birth choices’ – The Guardian

I am interested in the safety of homebirths because my wife and I seriously considered it for our first child and will certainly revisit the option again.  It is something that I perceive as safe, though NICE guidelines (2007) point to a lack of research comparing the safety of homebirths to hospital births. For those who don’t know, homebirths are offered by the National Health Service (NHS; United Kingdom), and though used infrequently ( 2.8% of births in 2007), they are a part of “normal” care options. Because they are less expensive than hospital births, they are promoted to varying degrees across the UK, but there is on-going debate regarding their safety. Spend a little time trawling the internet for “discussions” of homebirth safety and you’ll quickly have some idea of what a hot-button topic this can be.


Returning to the more alarming headlines, there were two things that immediately concerned me.  First, terms like “higher risk” are subjective, and typically based on relative measures of risk, rather than absolute measures. There is a great overview of this problem here, but in a nutshell, a large relative risk (e.g. “People who eat green jelly beans have five times the risk of disease X”) can be the ratio of two tiny absolute risks (e.g. five in a million jelly bean eaters get disease X, which only occurs one in a million times among non jelly bean eaters).  Second, when news media talk about risk, they rarely talk about all of the possible outcomes that could be associated with a particular exposure.  For example, it’s entirely reasonable that jelly beans could increase your risk of disease X while simultaneously reducing your risk of disease Y. Any rational decision about eating jelly beans would clearly benefit from seeing the larger picture, which is rarely provided in a news story.


To help understand these conflicting headlines I went to the source – Perinatal and maternal outcomes by planned place of birth for healthy women with low risk pregnancies: the Birthplace in England national prospective cohort study, published by the British Medical Journal. The study’s purpose was “To compare perinatal outcomes, maternal outcomes, and interventions in labour by planned place of birth at the start of care in labour for women with low risk pregnancies.” The specific finding that some headlines focused on was that for nulliparous pregnancies (i.e. women giving birth to their first child) “the odds of the primary outcome were higher for planned home births (adjusted odds ratio 1.75, 95% CI 1.07 2.86)”, while there were no differences among multiparous women. What do we need to know to interpret these results and make informed decisions for ourselves?


What was the primary outcome?

It was a “composite outcome” of perinatal mortality and intrapartum related neonatal morbidities. Study participants were counted as a “primary outcome” if they experienced any of the following events: stillbirth after start of care in labour, early neonatal death, neonatal encephalopathy, meconium aspiration syndrome, brachial plexus injury, fractured humerus, and fractured clavicle. It is important to note that all of these outcomes are very bad, but they are not equally bad. This was noted by the authors as a weakness, but necessary because of how rare the individual events are. Even when combined like this, the total number of primary outcome events was 250 out of 63,827, or 4.3 events per 1000 births – giving birth in the UK is very safe. The breakdown of the primary outcome is in table 8.1 here (75% were due to neonatal encephalopathy or meconium aspiration syndrome; 13% [32/250 events] were due to still birth or early neonatal death).


What are the absolute risk differences?

As previously noted, the adjusted odds ratio comparing the odds of the primary outcome occurring in a planned homebirth versus a planned obstetric birth was 1.75. Because the outcome is rare, the odds ratio is basically equivalent to the risk ratio, so we can say that there was a 75% increase in the risk, while others might might take some liberties and say that the risk is “almost doubled” etc. But what are the absolute risks? Among nulliparous planned homebirths, there were 9.5 primary outcomes per 1000 births; for planned obstetric births there were 5.3 per 1000. So the absolute risk difference is about 4 events per 1000 births. Another way to think of it is to look at the population attributable risk, which in this case is about 11.5%. It means that if (a big if) a planned home birth among nulliparous pregnancies was a cause of the primary outcome (not simply a risk factor), then eliminating all planned home births would reduced the incidence of the primary outcome by 11.5%. The remaining 88.5% of primary outcomes would be due to something other than planned home birth. Each individual must make up their own mind about the level of risk they are willing to accept, and looking at the numbers in different ways can help with this.


What were the estimates of risk adjusted for?

The basic goal of adjustment is to help us make more valid causal inferences from statistical associations (such as an odds ratio). This analysis adjusted for maternal age, ethnicity, English fluency, BMI in pregnancy, marital status, deprivation, parity, and gestational age. Adjustment for these factors is only important if they are associated with the exposure (planned home birth). Looking at table 1 from the paper, we can see that women planning a homebirth (compared to women planning obstetric births) tended to be older, from less-deprived areas, white, married, and fluent in English. With the exception of being a bit older, these factors are associated with better birth outcomes in the UK.  Thus failure to adjust for most of these factors would make a planned homebirth look safer than it actually is – resulting in an estimated odds ratio that is biased towards the null (no difference between obstetric and homebirths). Looking at the paper’s figures, the impact of adjustment on the estimated associations was very small. Despite this, we can never rule out that there are unaccounted for factors that explain the apparent relationship between planning a homebirth and experiencing the primary outcome.


How does homebirth relate to other outcomes considered in this research?

So far we have only discussed the primary outcome, but what about other outcomes included in the study? The study found that planned obstetric births were much more likely to have complicating conditions identified at the start of labour (19.5% versus 5.1% of homebirths; these are listed at the bottom of table 1). Next, regarding interventions during pregnancy, there were striking differences among birth settings. Key differences are summarized in the table below. Planned homebirths were more likely to have a “normal birth” (defined as a birth without induction, epidural or spinal analgesia, general anaesthesia, forceps or ventouse delivery, c-section, or episiotomy) than planned obstetric births (87.9% versus 57.6% among all women; 89% vs. 66% among births with no complications at the start of labour). Among planned obstetric births, 25.6% of women did not initiate breastfeeding, while this was true for only 11.5% of planned homebirths (table 8.4 here). Keep in mind that there were no differences in the primary outcomes in the entire population, despite these other differences. For women prioritising a “natural” birth, it is considerably less likely to happen if planning an obstetric birth. 



Freestanding midwife unit

Alongside midwife unit

Spontaneous vertex birth [“Normal”]





Vaginal breech birth





Ventouse delivery





Forceps delivery





Intrapartum caesarean section [“C-section”]





Syntocinon augmentation





Immersion in water for pain relief





Epidural or spinal analgesia





General anaesthesia





No active management of 3rd stage










* When compared to a planned obstetric birth, we would expect this difference to be explained by chance alone less than 5% of the time


What is the impact of defining the exposure as “planned place of birth” instead of “actual place of birth”?

The authors note this was a strength of the study, though at first glance this may seem somewhat at odds with a primary outcome “designed to capture outcomes that may be related to the quality of intrapartum [during birth] care…” However, focusing on planned place of birth prevents the authors from inadvertently stacking the deck against obstetric deliveries. This could happen because many women who plan a birth outside of the hospital wind up in the hospital to give birth anyway. This was certainly true in this study, which found that among all women planning a home birth, 21% of them wound up transferring to the hospital immediately before (14.2%) or after delivery (6.2%). These figures were notably higher among women giving birth for the first time, with 45% of them transferring into the hospital. Giving figures by actual place of birth would make hospitals appear riskier if those transfers were disproportionately due to complications during labour that in turn increased the risk of the primary outcome.

Even though this was an appropriate thing to do, we still need to interpret the results in light of this definition of exposure. However, the published research paper (or supplementary data) doesn’t provide outcome information specifically for women planning a homebirth who wound up transferring to hospital, which is unfortunate. This is actually another reason why it was important to look at planned birth place rather than actual birth place, i.e. the option to transfer into an obstetric unit is an integral part any safe homebirth service – and the entire process must be judged as a whole. Regardless, some important questions remain unanswered:

How many of the planned homebirths that had complications or interventions during labour actually occurred in an obstetric unit? Maybe the differences between homebirth and obstetric units with regards to intrapartum complications and interventions are even larger than apparent in this study?

What are the figures for nulliparous planned homebirths that didn’t need to transfer? Do women who transfer into obstetric care have outcomes and interventions that are similar to planned obstetric births? Are they worse?

Is it possible that a disproportionate number of outcomes among planned homebirths were due to events that occurred during obstetric care after transfer? This is a valid question, but seems unlikely since obstetric birth is very safe according to this study.

Conversely, one might also ask how many poor outcomes were avoided among planned homebirths by transferring into the obstetric unit, but in some ways this is a moot question because there will never be NHS provided homebirth without the ability to transfer into obstetric care.



First I would like to make it clear that this was a very well conducted study, and any apparent criticisms are really just a function of how difficult it is to answer these kinds of questions (is homebirth safe?) in a scientifically robust way. I also think it’s important, in light of how the research has been misrepresented in some places, to quote the authors’ conclusions before adding my own thoughts:

Our results support a policy of offering healthy nulliparous and multiparous women with low risk pregnancies a choice of birth setting. Adverse perinatal outcomes are uncommon in all settings, while interventions during labour and birth are much less common for births planned in non-obstetric unit settings. For nulliparous women, there is some evidence that planning birth at home is associated with a higher risk of an adverse perinatal outcome. A substantial proportion of women having their first baby who plan to give birth in a non-obstetric unit setting are transferred to an obstetric unit.

I think this is a very reasonable conclusion and summary of their results.

One thing that stood out for me were the considerable differences in maternal outcomes and complications among planned birth places, which contrasted with the relatively small differences with respect to the primary outcome. Should this lead us to question the benefit of higher levels of intervention?

Another was how different the first births were overall (which is not surprising as some of my own research concerns firstborns and obesity risk). There were more bad outcomes among nulliparous pregnancies regardless of the place of birth (5.3/1000 events vs. 3.1/1000 for multiparous pregnancies). Furthermore, first births planned at home were much more likely to transfer to obstetric care than other births (45% vs. 12%). Also, the level of intervention seen in nulliparous homebirths (much of which presumably takes place after transfer) was much higher than multiparous homebirths (see table 8.6 here). So what it is about first births that makes them different? While there is certainly a biological element at play (giving birth is, on average, biologically easier after the first one), what seems overlooked is the impact of uncertainly – the unknown. I am a father of one, and I can assure you that I will be much less freaked out about the next birth. I think my wife would say the same. Stress, fear, etc. – these things cannot be good for healthy birth outcomes, and perhaps this helps explain some of the increased risk of first births. And what about health care providers? Do they approach first births differently? Do they approach homebirths differently? Not all midwives have the same level of training and experience with homebirths. Not all obstetricians are equally supportive of homebirth. Could apprehension about homebirth among care providers interact with the natural, parental stress and fear associated with first births to create a particularly risky situation? If so, it seems to me that the solution, at least partly, lies in more training and normalizing of the homebirth experience – not simply their elimination as an option for first time mums.

Comments welcome – Rules of Engagement here (disclaimer).



8 Responses to Are homebirths really risky?

  1. Congratulations on your first post.

    A few comments:

    1. To the extent that the Birthplace Study identifies a subgroup in which homebirth may be as safe as hospital birth, that subgroup is “women who can be relied upon not to experience any complication of any kind.” In other words, homebirth is safe if nothing goes wrong. If there is any chance of anything going wrong, homebirth is not safe.

    You wrote, “So what it is about first births that makes them different? While there is certainly a biological element at play (giving birth is, on average, biologically easier after the first one), what seems overlooked is the impact of uncertainly – the unknown.”

    It’s simpler than that. The first birth serves as an additional layer of screening. Certain serious complications are unpredictable, but if they happen in the first pregnancy, they are more likely to happen in the second pregnancy.

    So, for example, we don’t know which first time mothers are going to experience shoulder dystocias, so it is difficult to screen them out of the group of first time mothers. If a woman has a shoulder dystocia in a previous pregnancy, she is at much greater than average risk for having one in the second pregnancy. Screening for complications in an earlier pregnancy in addition to complications in the current pregnancy makes it even more likely that women in the homebirth group are those “who can be relied upon not to experience any complication of any kind.”

    2. It is critical to keep in mind that the criteria in the study are far more rigorous than the real life criteria for homebirth. Therefore, the results represent the theoretical perfect outcome, not a real world outcome. Even in this best case scenario, homebirth still increases the risk of death in the event of complications.

    Complications are going to be more common in the real world group of women who actually have homebirths and did not have to meet the more rigorous exclusion criteria of the study. The bottom line is that homebirth increases the risk of perinatal death, almost certainly far more than the study demonstrated.

    3. Are homebirths risky?

    That’s a subjective assessment, but the results tell us objectively that homebirth is not as safe as hospital birth. Each woman will have to decide for herself if the increased risk is acceptable, but there is no question that homebirth increases the risk of perinatal death.

    • Holly says:

      Wow, but then, what can one really expect from Ms Tuteur? Way to twist the results. I’m afraid you are flat out WRONG. This study conclusively demonstrates that for LOW RISK, multiparous women, planning a home birth is the safest option for both mum and baby and that for LOW RISK primiparous women, planning a home birth is still the safest place for them with only a small risk to their babies.

      Freestanding and alongside midwifery units were shown to be safer than obstetric units for all low risk women and babies, regardless of parity.

      Maybe you should look at the actual results before stamping your twisted conclusions upon them?

      You are right, not all home births in general are considered fit for the “low risk” label, but this study did not set out to analyse this subgroup. More research needs to be done on place of birth by risk factor in order for all women to make fully informed decisions about where they give birth, however, to dismiss this study on the basis that it is not “real world” is ignorant in the extreme. The women and babies in this study all exist, they ARE a “real world” sample, they have simply been selected on a basis which allows the researchers to draw useful and meaningful conclusions that will enable the majority of women in the UK to make informed decisions about where they plan to have their babies.

  2. Holly says:

    Thank you for a very comprehensive, honest and positive assessment of the results of the Birthplace Report.

  3. Rachel Cline says:

    Thank you so much for this detailed analysis of the study itself. It is so interesting how the different elements of the media interpret a story and the headlines that spin from it, and this happens so often and polarizes people in this very sensitive topic are, thank you for giving us a more grounded and rounded view point.

  4. Celia Grigg says:

    Thank you for your comments on the Birth Place England research. I would like to ask you a question regarding the powering of statistics. The authors state that the study is not powered to detect statistically significant differences in stillbirth or neonatal death, due to their rarity and the sample size of the current study. If this study is not powered for still birth or neonatal death, how does adding them to the composite outcomes give them power? Surely, if the study is not powered to detect these variables independently it is not powered for them as composite outcomes also?

    • dldahly says:

      Hi Celia. You are basically correct. The study is not powered to detect meaningful differences in stillbirths or neonatal deaths (since they are so rare), hence the use of a composite indicator to provide an outcome with more events, thus making differences between groups easier to detect.

      There isn’t really anything wrong with this per se – we just need to make sure that any statistical inferences we make about group-based differences (or lack thereof) are focused on the composite indicator, not the individual components. In other words, as long as we are talking about the “primary outcome”, we are ok.

  5. Our organisation along with statisticians, have crawled over the data in this study for one reason and one reason only; the loss or brain damage to a child is one of the most devastating things that can happen to a family. Families have the right to honest information.

    Yes, this is broadly a good study but there are a few flaws that almost invalidate one or two of the conclusions.

    Maternal outcomes.

    Of the many definitions of normal birth, NPEU chose one that excluded epidurals. Since many of the women would have specifically chosen obstetric birth because of the availability of epidurals, this vastly biased the outcomes against obstetric units. Their definition, unlike the NICE one, makes no reference to a healthy mother and baby. So the normal births could include a woman with a 4th degree tear, a catastrophic haemorrhage, PTSD and last but not least, a dead baby. No claim can be made that the normal birth rate defined in this study improves maternal outcomes.

    First time mums in MLUs

    Because Birthplace was looking at relatively rare outcomes, it was necessary to get as close to 100% of forms back during the period under study.

    The study states, I quote ‘to minimise non-response bias we aimed to include a minimum of 85% of eligible women in each participating unit or trusts.

    You expect, therefore, to see a table showing the results from units which returned 85% of the data fairly close to the front of the analysis summary. There is indeed such a table, it represents 73% of the data returns for the group in question, but it is relegated to page 213. (Table 59). It shows that for first time mums, birth in a freestanding midwifery unit is twice as risky and that risk is statistically significant. This increased risk for first time mums in midwifery units has been seen in several other studies. (1)

    If we then look at the remaining 27% of data – those where the units returned less than 85% of data forms – we get a rather bizarre and reverse result. Adverse outcomes are 56% safer in the Free standing midwifery units – a finding no other research study has ever concluded. Moreover, there are real reasons why FMU units that were less than conscientious might return fewer poor outcomes – simply because these women would have been transferred to the Obstetric unit and the data would be harder to collect. OU units who were less than switched on to the study might well have returned more adverse outcomes because they might have seen them as more important. Finally, and perhaps a rather sensitive issue, is that the jobs of midwives in freestanding units would depend on the outcome of the study and they were the ones filling in and returning the forms.

    NPEU pooled the data, and confidently asserted no extra risk for first time mums.

    We strongly disagree. The largest data set with the smallest confidence interval indicates an increase in risk. The other data is much smaller, we have no idea of the confidence interval and the results are diametrically opposed to the better data and strikingly unusual.

    Statistics isn’t the problem here. The problem lies with the quality of the data that was collected….and it looks a bit dodgy.

    What should they have done? We certainly believe that on the basis of this data, and given the concerns over data collection, that at very least you say, we don’t know for sure and state the results clearly and upfront. By asserting no risk, they are actually flying in the face of the greater probability that the >85% data is right. It may be an embarrassment for NPEU but it is better than betraying families.

    (1)Hodnett ED (2004) Home-like versus conventional institutional settings for birth
    (Cochrane Review) In: The Cochrane Library, Issue 3, 2-4. Chichester, UK, John
    Wiley & Sons Ltd

    Moster D, Terje Lie, Markestad T (2001) Neonatal mortality rates in communities
    with small maternity units compared with those having larger units British Journal
    of Obstetrics and Gynaecology 108: 904-909

    Gottvall K, Grunewald C, Waldenström U (2004) Safety of birth centre care:
    perinatal mortality over a 10-year period BJOG: an international journal of
    obstetrics and gynaecology 111: 71-78

    Waldenström U, Nilsson C-A, Winbladh B (1997) The Stockholm Birth Centre
    Trial: maternal and infant outcome British Journal of Obstetrics and Gynaecology
    104: 410-418

  6. Fredrik says:

    Dear Dr. Dahly,

    I read your comments on the Birthplace in England (BiE) report. I thought they were good, some things one might think about is adjusting for covariates; that certain factors are risk factors may be due to a specific setting (e.g. being well-educated in England may be advantageous giving birth at a hospital but not necessarily so in a home birth situation where one hasn’t to demand for care and attention) so how do one correct for these? Generally I’d say that one tend to just adjust for all factors additively which wont be correct if such a difference is the case. Also making age/BMI discrete as they did in BiE is not the best way to adjust for nonlinearites, one loose a lot of information.

    So I thought: what if I look at the raw figures and check if the comparisons are adjusted for multiple comparisons? I found that the figures of risk differed quite a lot, since the unadjusted risk is in fact adjusted for some sampling probability which makes it rather opaque.

    The overall test shows that the relative risks are much higher for obstetric units than for the other ones (i.e. I take “totally unadjusted RRs”, see R-script below):

    Linear Hypotheses:
    Estimate Std. Error z value Pr(>|z|)

    Along midw – Ob.unit == 0 -0.7818 0.1720 -4.545 |z|)

    Along midw – Ob.unit == 0 -0.06933 0.21342 -0.325 0.9800

    Free midw – Ob.unit == 0 -0.05847 0.24677 -0.237 0.9920

    Home – Ob.unit == 0 0.56618 0.21183 2.673 0.0216 * ,

    Which transforms to the RRs:

    Estimate lwr upr
    Along midw – Ob.unit 0.9330231 0.5631814 1.545740
    Free midw – Ob.unit 0.9432100 0.5261332 1.690912
    Home – Ob.unit 1.7615307 1.0672747 2.907396,

    These figures are much closer to the figures reported in Table 2:

    Home 1.76 (1.10 to 2.82)
    Freestanding midwifery unit 0.85 (0.49 to 1.48)
    Alongside midwifery unit 0.90 (0.53 to 1.54) ,

    but note that my confidence intervals all cover the reported ones; maybe these aren’t adjusted for multiple comparisons?

    On the other hand, considering the comparison of the outcomes of two different populations is indeed a very risky business as the Hormone Replacement Therapy by Womens Health Initiative and many other randomized controlled trials of epidemiological findings have shown. What if the nulliparous home birth population have other risk factors than the ones found in the overall comparison? What about the unmeasured risk factors?

    Interestingly, in Sweden it is not allowed to move pregnant mares 35 days before planned births (6 weeks if transport longer than 1 day) and not until more than 7 days after. ( paragraph “Dräktiga ston”, google translate works about ok).
    In my experience the transport to the hospital can be extremely stressful to both parents (and most likely to the child) and as you say stress is not good.

    Finally, I agree with the conclusion of yours and the article. And, as you say, that without the back-up from hospitals home birth would not be a very good idea.

    All the best,


    "Free midw","Along midw"))
    location<-relevel(location, ref=4)
    #makes Ob.unit reference
    offset(log(births)), family=poisson)
    anova(primo.glm, test="Chisq")


    # shouldn’t one spend the degrees of freedom
    # e.g. critical level 0.03 on
    # primary and 0.01 on second and third?
    # One way of doing multiple comparisons
    # adjustments for multiple hypotheses
    confint(sec.glht, level=0.99)


    trd.glht<-glht(trd.glm, linfct=mcp(location="Dunnett"))

    Funny quotes from Cosma Shalizi’s lecture notes :
    “Every time someone uses linear regression with the standard assumptions for inference and does not test whether the residuals are white noise, an angel loses its wings.”
    …and then…
    “Every time someone thoughtlessly uses regression for causal inference, an angel not only loses its wings, but is cast out of Heaven and falls in most extreme agony into the everlasting fire.”

    The data are of course not treated with linear regression but the underlying data should be made accessible for peers to scrutinize.

Leave a Reply

Your email address will not be published. Required fields are marked *

You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <strike> <strong>