Piaggio et al.

Reproductive Health 2018, 15(Suppl 1):97

https://doi.org/10.1186/s12978-018-0530-7

RESEARCH Open Access

Prevention of postpartum haemorrhage: a

distributional approach for analysis
Gilda Piaggio1*, José Ferreira de Carvalho2 and Fernando Althabe3
From 2nd International Conference on Maternal and Newborn Health: Translating Research Evidence to Practice
Belagavi, India. 26-27 March 2018

Abstract
Background: There is empirical evidence that measured postpartum blood loss has a lognormal distribution. This
feature can be used to analyze events of the type ‘blood loss greater than a certain cutoff point’ using a lognormal
approach, which takes into account all the quantitative observations, as opposed to dichotomizing the variable blood
loss volume into two categories. This lognormal approach uses all the information contained in the data and is
expected to provide more efficient estimates of proportions and relative risk when comparing treatments to prevent
postpartum haemorrhage. As a consequence, sample size can be reduced in clinical trials, while keeping the statistical
precision requirements.
Methods: The authors illustrate how a lognormal approach can be used in this situation, using data from a clinical trial
and the event ‘blood loss greater than 1000 mL’.
Results: Estimates of the proportions of this event for each treatment, and relative risks obtained with this
method are presented and compared with the standard estimates obtained by dichotomizing measured
blood loss volume. An example of how the blood loss distributions of two treatments can be compared is
also presented. Different scenarios of the sample size needed to compare two treatments or interventions
are presented to illustrate how with the lognormal approach the size of a clinical trial can be reduced.
Conclusions: A distributional approach for postpartum blood loss using the lognormal distribution fitted to
the data results in more precise estimates of risks of events and relative risks, compared to the use of
binomial proportions of events. It also results in reduced required sample size for clinical trials.
Trial registration: This paper reports a secondary analysis for a trial that was registered at clinicaltrials.gov
(NCT00781066).
Keywords: Postpartum blood loss, Postpartum haemorrhage, Severe postpartum haemorrhage, Lognormal distribution

Background the distribution. This has been the case with postpartum
The development of an adequate statistical analysis tech- blood loss, for which a logarithmic transformation has
nique to analyze a continuous variable depends on the been used to compare medians, [1] based on the obser-
knowledge of its distribution. Many variables in biology vation that the blood loss distribution is positively
and medicine follow the normal distribution and stand- skewed, as the lognormal distribution. Also, there are
ard statistical techniques can be applied to compare oftentimes physical or biological justifications for a vari-
means. However, when the distribution is not normal, able to have a specific distribution. The lognormal distri-
the standard statistical techniques are no longer appro- bution is a result of many independent small
priate and a transformation is often used to normalize multiplicative effects. It is a simple model that applies to
many problems such as body and tumor mass (weight)
* Correspondence: [email protected]
[2], and blood pressure [3]. Examination of simple histo-
1
Statistika Consultoria, Campinas, São Paulo, Brazil grams of blood loss reveals right skewed distributions
Full list of author information is available at the end of the article

© The Author(s). 2018 Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0
International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and
reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to
the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver
(http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.
Piaggio et al. Reproductive Health 2018, 15(Suppl 1):97 Page 62 of 126

that resemble the lognormal distribution, and more in- Once a lognormal distribution is considered to be a
depth and formal statistical analysis showed that indeed good fit to the data, the fit is visualized by plotting the
the lognormal distribution fits the blood loss distribution cumulative distribution function for the observed data
very well. [4] (‘empirical cumulative distribution function’) together
In the case of postpartum blood loss, there is interest with the fitted lognormal cumulative distribution func-
to compare events of the type ‘blood loss beyond a cer- tion, or, alternatively, its complement, denoted here as
tain cutoff point’ because the loss of large amounts of the survival function. The “survival” function gives the
blood postpartum can lead to severe maternal morbidity probability of having blood loss MORE than a particular
and mortality [5, 6]. Therefore there has been a concern value. It is R(v) = 1-F(v), where F(v) is the cumulative
to find efficient treatments or interventions to prevent probability function.
postpartum haemorrhage (PPH), defined as blood loss of The probability of an event of the type ‘blood loss
500 ml or more within 24 h after birth, and severe PPH greater than a cutoff point’ is just the survival function
(sPPH) as blood loss of 1000 ml or more within 24 h at the cutoff point. For example, the proportion of sPPH
after birth [7]. is just the survival distribution at the point 1000.
Measured blood loss is thus categorized in two cat- Comparison of proportions between treatments and
egories, by means of an indicator variable of blood loss computation of relative risks with confidence intervals are
greater than a certain cutoff point. The estimation ap- done using bootstrap techniques. We generated one thou-
proach used so far has been to compute the sample pro- sand bootstrap samples for each treatment. The estimates
portion of women with blood loss equal to or above the of the proportions of sPPH and PPH are then computed
cutoff point, or a binomial proportion. However, the for each bootstrap sample, for each treatment. [13] The
categorization of a dependent variable results in a loss of two bootstrap samples tables are matched by row (sample)
power to detect true effects, which is substantial if the and the relative risks computed. From the distribution of
distribution is highly skewed and if the categorization is the 1000 bootstrapped relative risks, the 95% confidence
done in few categories, or both [8]. interval can be obtained from the 2.5 and 97.5% percen-
We have shown empirically, elsewhere [4], that the dis- tiles of the distribution of the 1000 samples.
tribution of postpartum blood loss volume is lognormal, To illustrate the gain in precision for this data, we esti-
using data provided by the authors from three trials that mated the proportions and relative risk in the standard
compared two drugs [1, 9] or two management proce- way, with 95% confidence intervals, denoted the bino-
dures for the third stage of labour [10], and one observa- mial approach. For the proportions, we calculated the
tional study [11]. We used this finding to propose an width of the 95% confidence intervals for both ap-
analysis approach based on the lognormal distribution, proaches, the lognormal approach and the binomial ap-
resulting in more efficient estimates of proportions and proach, and calculated their ratio as a quantification of
relative risk and in a reduction of the sample size needed the gain in precision. For the relative risk we applied a
in clinical trials that compare proportions between treat- similar procedure but on the relative scale.
ments [4]. In this paper we illustrate how this approach We also illustrate tests of hypothesis using the two ap-
(denoted ‘the lognormal approach’) can be used to analyze proaches. For the test of equality of the sPPH propor-
data from one of these trials, the Althabe et al. trial [1]. tions between the two treatments, using the binomial
approach, we report the Pearson chi-square statistic and
Methods p-value. To test the equality of the distributions for the
Descriptive histograms by treatment are constructed to two treatments using the lognormal fits, we proceed se-
have a first view of the distributions. quentially: first we test the model with equal scale pa-
The lognormal approach that we propose uses mea- rameters versus the full model (both location and scale
sured blood loss observations without categorizing this parameters possibly different). If the null hypothesis with
variable. It consists of the following steps: equal scale parameters is not rejected, we test the model
The procedure starts by fitting a three-parameter log- with equal location parameters against the model with
normal distribution [12] to the data. The parameters of possible different location parameters and conclude
the lognormal distribution are estimated by maximum about whether the distributions differ at, say, 5% level of
likelihood. significance.
Goodness of fit is assessed using probabilistic plots, All the computations for fitting distributions and
consisting in plotting the quantiles of the fitted lognor- obtaining estimates were done with JMP® 13 soft-
mal distribution against the observed blood loss values. ware. [14]
If the fit is good, then the points will fall on a straight Computations for sample size calculations were
line. The probabilistic plot is also used to detect pres- done with SAS® software version 9.4 (PROC POWER
ence of outliers and to assess the quality of the data. procedure) [15].
Piaggio et al. Reproductive Health 2018, 15(Suppl 1):97 Page 63 of 126

Results Table 1 Estimated parameters for the fitted threshold

Descriptive histograms of blood loss volume lognormal distribution for the Althabe et al. trial [1]
In Fig. 1 we show histograms of frequency distributions (the three parameters are denoted location, scale and
of blood loss for the Althabe et al. trial (1), by treatment. threshold parameters); treatments are hands-off and
The distributions are right-skewed, but from the histo- control cord traction (CCT)
grams we cannot specify the statistical distribution ori- Treatment Parameter Estimate Std Error 95% CI
ginating the data. Hands-off location 5.57 0.141 5.30 to 5.85
scale 0.72 0.101 0.52 to 0.92
Fitting a lognormal distribution threshold 55.14 24.474 7.18 to 103.11
A three-parameter lognormal distribution (threshold CCT location 5.37 0.132 5.11 to 5.63
lognormal, abbreviated as THLN) was fitted to the data
scale 0.80 0.101 0.60 to 1.00
by maximum likelihood. The third parameter was added
because it improved the fit compared to the two- threshold 62.88 16.414 30.71 to 95.05
parameter lognormal distribution. The estimated param-
eters are shown in Table 1, with their standard errors three-parameter lognormal distribution (THLN), in a
and the 95% confidence intervals. superimposed red line, with 95% confidence band (red
The goodness of fit of the THLN distribution to the area). In Fig. 3 we can see, for example, that the prob-
data can be visualized in the lognormal probability plot ability of having 500 mL or more of blood loss is about
of Fig. 2. The probabilities from the fitted lognormal dis- 0.20. The lognormal distribution fits the data points very
tribution and the data points are on a straight line for well. The 95% confidence intervals from the lognormal
values above 50 mL, thereby showing that the fit of the fit are narrower than the ones for the binomial
THLN distribution to the data is very good above 50 mL estimates.
and providing evidence that the lognormal distribution
is appropriate to the blood loss volume distribution. No Estimates of proportions and relative risk
outliers were detected. Only one treatment is shown in Estimation of proportions using number of events di-
Fig. 2 for illustration purposes, because the two treat- vided by the total number in each treatment will be de-
ments had a very similar behaviour. noted the binomial approach. Estimation of proportions
In Fig. 3 we show the survival function for the ‘hands- using the survival function at the relevant cut-off point
off’ treatment (n = 98), where the data points are repre- will be denoted by the lognormal approach. We can see
sented by black dots and the pointwise 95% confidence in Fig. 3 that the number of black dots above 1000 mL
intervals by blue lines. Figure 3 also shows the fit of a is 5, resulting in an estimated proportion by the

Fig. 1 Frequency distribution of blood loss volume (mL) by treatment, Althabe et al. trial [1]
Piaggio et al. Reproductive Health 2018, 15(Suppl 1):97 Page 64 of 126

Fig. 2 Probabilistic plot of the THLN fitted curve (red line) with pointwise 95% confidence band (red area), showing the data points (black dots)
with their 95% confidence intervals (blue lines), Althabe et al. trial, hands-off treatment [1]

Fig. 3 THLN fitted curve (red line) with pointwise 95% confidence band (red area), the empirical survival function (black dots) with the pointwise
95% confidence intervals (blue lines), Althabe et al. trial, hands-off treatment [1]
Piaggio et al. Reproductive Health 2018, 15(Suppl 1):97 Page 65 of 126

binomial approach, of 5/98 = 0.051, or 5.1%, for the Table 3 Relative risk with 95% confidence intervals for the
hands-off treatment. The estimate of the proportion of binomial and lognormal approaches, and gain in precision,
sPPH, based on the fitted survival function, is 0.038 (0. Althabe et al. trial [1]
017 to 0.078). It is interesting to note that, of the five Approach RR (95% CI) Ratio upper/lower Log scale width
CCT vs Hands-off limit of the 95% CI ratio lognormal vs
points in excess of 1000, three are close to the cut-off
binomial (%)
point. They could easily have slipped, by chance, to close
Binomial 0.58 (0.14 to 2.37) 16.9 –
values to the left of the cut-off, thereby sharply lowering
the binomial estimate. Such a change would barely affect Lognormal 0.86 (0.22 to 2.62) 11.9 70.4
the lognormal estimate, since the fitted curve closely fol-
lows the entire cumulative empirical distribution and The lognormal tests of hypothesis that the two distri-
takes into account all the data. butions are the same is shown in Table 4. The compari-
Table 2 shows the estimated proportions of sPPH for son of the scale parameters shows that they are not
the Althabe et al. trial by the binomial approach, as re- significantly different, as the p-value equals 0.7381;
ported in the published trial results [1] and by the log- therefore there is no evidence that the scale parameters
normal approach, per treatment group. of the two treatments are different. The location param-
As can be appreciated in Fig. 3, Table 2 also shows that eters are not significantly different either, as the p-value
the 95% confidence intervals for the lognormal approach equals 0.1506. Hence we conclude that there is no evi-
are narrower than the ones for the binomial approach. dence against the equality of the distributions.
For the hands-off treatment, for example, the width of The significance level for the final test, comparing the
the binomial estimate confidence interval is 0.092, location parameters, is about 0.15 for the lognormal ap-
whereas that of the lognormal estimate confidence inter- proach. This can be compared to the significance level
val is 0.062, about two thirds of the former. The 95% of the binomial test, 0.44. The lognormal approach
confidence intervals width ratio is 67% and 75% respect- seems more sensitive.
ively for the hands-off treatment and for the controlled Figure 4 shows the two distributions side by side on a
cord traction (CCT) treatment. lognormal probability plot. The joint confidence inter-
For the lognormal distribution, the relative risk, shown vals overlap entirely, suggesting that the two distribu-
in Table 3, with the 95% confidence intervals, is esti- tions are very similar.
mated by the bootstrap technique, using 1000 bootstrap
samples. Note that the confidence intervals for the RR Sample size: an example
are wide because this is a small trial that was designed We present as an example different scenarios of sample
to compare median blood loss as the main outcome. size calculation using the two approaches, the binomial
From the comparison of the 95% confidence limits for and the lognormal one.
both approaches, we obtained a log-scale width ratio of For the binomial approach, we assume that the pro-
70.4% for the lognormal approach in relation to the bi- portion of sPPH is 0.015, 0.02 or 0.025 in the current
nomial approach, so that the gain in precision when treatment. We also assume that a new preventive ther-
using the lognormal instead of the binomial approach, is apy is considered worthwhile if the relative risk of sPPH
about 30%. of the new therapy with respect to the current one is no
larger than 0.70 or 0.80. We calculated the sample size
for the binomial response based on the likelihood ratio
Tests of hypothesis chi-square one-sided test for the relative risk statistic.
The test of equality of sPPH proportions by the binomial For the lognormal approach, we used the well known re-
method can be based on a 2 × 2 table, giving the Pearson sult that if a variable has a lognormal distribution, its loga-
chi-square = 0.591 with p-value = 0.4440. rithm has a normal distribution. Therefore, our sample

Table 2 Estimated proportions of sPPH for the Althabe et al. trial [1] by the binomial and by the lognormal approaches, per
treatment group; treatments are hands-off and control cord traction (CCT)
Trial Treatment n/N Proportion 95% CI Width of the 95% CI Width ratio lognormal vs binomial (%)
Binomial Hands-off 5/98 0.051 (0.022 to 0.114) 0.092 –
CCT 3/101 0.030 (0.010 to 0.084) 0.074 –
Lognormal Hands-off – 0.038 (0.017 to 0.078) 0.062 67
CCT – 0.033 (0.014 to 0.069) 0.055 75
Piaggio et al. Reproductive Health 2018, 15(Suppl 1):97 Page 66 of 126

Table 4 Results of lognormal tests of equality of distributions of with TWOSAMPLEMEANS was used for the
the two treatments, Althabe et al. trial [1] computations.
Models compared Likelihood ratio Chi-square DF p-value The total sample sizes for the two approaches, for a
No effect vs. location 2.066 1 0.15 power of 80%, in a one-sided 5% significance test, are
Location vs. location and scale 0.112 1 0.74 shown in Table 5. The difference in required sample size
is enormous, as expected, because the lognormal ap-
proach is using more of the information in the sample.
size computations were based on the transformation of
the volumes to their logarithms. For a scenario consisting
of a given proportion of sPPH, say p, and a relative risk Discussion
RR = p2/p1, we computed the corresponding risk of the We have illustrated how to apply an analysis technique
competing treatment, p2. For the standard deviation on for blood loss volume data based on the lognormal dis-
the log scale, we used the scale parameter obtained from tribution, without categorizing this response variable, for
the analysis of three clinical trials, that was in all cases a small trial [1], verifying first that the blood loss volume
close to s = 0.7. [4] The values of the mean m (on the log indeed follows a lognormal distribution. Using data from
scale) can be readily computed for each value of p and for two large trials [9, 10], the same pattern was found, that
each scenario, from an equation derived from the propor- we reported elsewhere [4]. For these two large trials, the
tion p of sPPH: fit was also very good. We have also fitted a lognormal
distribution to blood loss data using the reported per-
logð1000Þ−m centiles from an observational study [11]. In all cases,
¼ z1−p we have found empirical evidence that the blood loss
s
volume has a lognormal distribution, and can be de-
where z1-p is the (1-p) quantile (or (1-p)× 100% percent- scribed by a variant of this family of distributions, the
ile) of the standard normal distribution, taking values p1 threshold three-parameter lognormal distribution [12].
and p2 for a particular scenario. With the two computed A lognormal distribution with its specific parameters
means, derived from p1 and p2 using the equation above, characterizes several physical and biological phenomena
and the (fixed) standard deviation, together with the [16], and can be described by means of a physical model
power requirement, the computation of the sample size as a multiplicative sequence of losses.
is straightforward, done as a comparison of means of the Furthermore, the estimated location and scale parame-
log-transformed variable blood loss. SAS PROC POWER ters from all four of the studies analyzed were very

Fig. 4 Distributions of blood loss for the two treatments, Hands-off (red) and CCT (blue), side by side on a lognormal probability plot, showing
data points (dots), fitted lognormal lines (continued full lines), and 95% confidence bands (shaded areas), Althabe et al. trial [1]
Piaggio et al. Reproductive Health 2018, 15(Suppl 1):97 Page 67 of 126

Table 5 Sample size required for the two different approaches, binomial and lognormal, under different scenarios, for a one-sided
5% significance test with 80% power for relative risk (RR)
Scenario Assumed sPPH rate for control (%) RR Total sample size with the binomial approach Total sample size with the lognormal approach
1 1.5 0.70 15,294 1178
2 1.5 0.80 36,522 3068
3 2.0 0.70 11,422 1080
4 2.0 0.80 27,266 2814
5 2.5 0.70 9098 1002
6 2.5 0.80 21,714 2618

similar [4]. The studies were conducted in different between the simplicity offered by the binomial approach
places and times, suggesting that the lognormal distribu- and the possibility of reducing the size of a trial.
tion fits postpartum blood loss data universally. The sta- Similar methods described in this paper can be used
bility of the parameters found in the analysis of blood with other variables having the lognormal distribution,
loss data across studies may well be a characteristic of like blood pressure [3] and estimation of hypertension.
postpartum blood loss that can be further explored. The reason why this approach has not been used in the
The available analysis technique to compare proportions past, is that it is computer intensive. Nowadays, with im-
of an event of the type ‘blood loss above a certain cut-off provement in computer power, this is not a problem, al-
point’ between treatments or interventions to prevent this though the complexity of fitting a lognormal distribution
event, has been to estimate the two binomial proportions and calculating relative risk’s confidence intervals by
of sPPH. The categorization of blood loss volume in two bootstrapping requires statistical expertise.
categories entails loss of information contained in object- As an additional bonus to the use of the lognormal ap-
ively measured weight or volume data, with a resulting loss proach based on fitting a lognormal distribution to
of power in tests of hypothesis and a decrease in the preci- blood loss data, it is possible to test other hypotheses of
sion of the estimates of proportions and relative risks [8]. interest, such as the equality of medians or any other
This fact, together with the low prevalence of rare events percentile, or even compare the entire distributions be-
when the cut-off point is a high value of blood loss volume, tween two treatments or interventions.
for example 1000 mL, results in very large size of trials
needed to compare this event between treatments or Conclusions
interventions. We illustrated how a lognormal approach based on fit-
We proposed a lognormal approach of analysis of post- ting a lognormal distribution to the data can be applied
partum haemorrhage trials aiming to compare events of to measured blood loss volume data of a trial. We found
the type ‘blood loss greater than a certain cutoff point’ be- that the precision of the estimates of proportions of the
tween treatments. [4] We illustrate here this approach, con- event ‘blood loss greater than 1000 mL’ and its compari-
sisting of fitting a lognormal type of distribution to blood son between treatments improved compared to the
loss data, which involves estimating the parameters that de- standard methods based on dichotomizing the blood loss
fine the distribution. Once the distribution function of variable. We also illustrate how the lognormal approach
blood loss volume, or its complement, the survival function, can be used to compare the distribution parameters for
is defined by its parameters, the proportion of sPPH, for ex- two treatments. When analyzing data using this lognor-
ample, is just the survival distribution at the point 1000. To mal approach, sample size of trials can be reduced.
compare treatments using relative risk, we estimated its
Acknowledgements
confidence interval with bootstrap techniques. The authors thank the reviewers for valuable comments.
An application of using the lognormal model for the
distribution of the blood loss volume is a substantial re- Funding
Publication charges for this supplement were funded by the University of
duction of sample size of clinical trials, while keeping British Columbia PRE-EMPT (Pre-eclampsia/Eclampsia, Monitoring, Prevention
the statistical power and precision requirements. Using and Treatment) initiative supported by the Bill & Melinda Gates Foundation.
the lognormal approach that we propose, based on fit-
Availability of data and materials
ting a lognormal distribution to the postpartum blood The datasets used and/or analysed during the current study are available
loss data, the objectives of a trial can be attained with from the corresponding author on reasonable request.
smaller sample sizes and reduced cost, through an im-
About this supplement
provement in the efficiency of the estimation methods. This article has been published as part of Reproductive Health Volume 15
With the lognormal approach, there is a trade-off Supplement 1, 2018: Improving pregnancy outcomes - Proceedings of the
Piaggio et al. Reproductive Health 2018, 15(Suppl 1):97 Page 68 of 126

2nd International Conference on Maternal and Newborn Health: Translating 13. Efron B, Tibshirani R. Bootstrap methods for standard errors, confidence
Research Evidence to Practice. The full contents of the supplement will be intervals, and other measures of statistical accuracy. Statist Sci. 1986;1(1):54–75.
available online at https://reproductive-health-journal.biomedcentral.com/ 14. JMP®, Version 13. SAS Institute Inc., Cary, NC, 2017.
articles/supplements/volume-15-supplement-1. 15. SAS/STAT, software. Version 9.4 for Linux, SAS Institute Inc. ed. Inc. SI, editor.
Cary, NC, USA.2012.
Authors contributions 16. Limpert E, Stahel WA, Abbt M. Log-normal distributions across the sciences:
GP: conceptualization, writing (original draft, editing and review). JFC: keys and clues. Bioscience. 2001;51(5):341–52.
conceptualization, statistical analysis, writing (original draft, editing and
review). FA: writing (editing and review), provided the data. All authors read
and approved the final manuscript.

Ethics approval and consent to participate

Not applicable.

Competing interests
The authors declare not to have any competing interests.

Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in
published maps and institutional affiliations.

Author details
1
Statistika Consultoria, Campinas, São Paulo, Brazil. 2Statistika Consultoria,
Campinas, São Paulo, Brazil. 3Institute for Clinical Effectiveness and Health
Policy (IECS-CONICET), Department of Mother & Child Health Research,
Buenos Aires, Argentina.

Published: 22 June 2018

References
1. Althabe F, Aleman A, Tomasso G, Gibbons L, Vitureira G, Belizan JM, et al. A
pilot randomized controlled trial of controlled cord traction to reduce
postpartum blood loss. International journal of gynaecology and obstetrics:
the official organ of the International Federation of Gynaecology and
Obstetrics. 2009;107(1):4–7. Epub 2009/06/23
2. Spratt JS Jr. The lognormal frequency distribution and human cancer. J Surg
Res. 1969;9(3):151–7. Epub 1969/03/01
3. Makuch RW, Freeman DH Jr, Johnson MF. Justification for the lognormal
distribution as a model for blood pressure. J Chronic Dis. 1979;32(3):245–50.
Epub 1979/01/01
4. Carvalho JF, Piaggio G, Wojdyla D, Widmer M, Gülmezoglu AM. Distribution
of postpartum blood loss volume: modeling, estimation and application to
clinical trials. 2018; Submitted.
5. Say L, Chou D, Gemmill A, Tuncalp O, Moller AB, Daniels J, et al. Global
causes of maternal death: a WHO systematic analysis. Lancet Glob Health.
2014;2(6):e323–33. Epub 2014/08/12
6. Khan KS, Wojdyla D, Say L, Gulmezoglu AM, Van Look PF. WHO analysis of
causes of maternal death: a systematic review. Lancet. 2006;367(9516):1066–
74. Epub 2006/04/04
7. WHO. Recommendations for the prevention and treatment of postpartum
haemorrhage. World Health Organization. 2012.
8. Taylor AB, West SG, Aiken LS. Loss of power in Logistic,Ordinal logistic, and
Probit regression when an outcome variable is coarsely categorized. Educ
Psychol Meas. 2006;66:228–39.
9. Gulmezoglu AM, Lumbiganon P, Landoulsi S, Widmer M, Abdel-Aleem H,
Festin M, et al. Active management of the third stage of labour with and
without controlled cord traction: a randomised, controlled, non-inferiority
trial. Lancet. 2012;379(9827):1721–7. Epub 2012/03/09
10. Gulmezoglu AM, Villar J, Ngoc NT, Piaggio G, Carroli G, Adetoro L, et al.
WHO multicentre randomised trial of misoprostol in the management of
the third stage of labour. Lancet. 2001;358(9283):689–95. Epub 2001/09/12
11. Bamberg C, Niepraschk-von Dollen K, Mickley L, Henkelmann A, Hinkson L,
Kaufner L, et al. Evaluation of measured postpartum blood loss after vaginal
delivery using a collector bag in relation to postpartum hemorrhage
management strategies: a prospective observational study. J Perinat Med.
2016;44(4):433–9. Epub 2015/09/10
12. Meeker WQ, Escobar LA. Statistical methods for reliability data: John Wiley &
Sons. Inc.; 1998.