Magnesium sulphate for prevention of eclampsia: are intramuscular and intravenous regimens equivalent? A population pharmacokinetic study


Correspondence: Dr T Easterling, Department of Obstetrics and Gynecology, University of Washington, 1959 Pacific NE, Box 356460, Seattle, WA 98195, USA. Email



To compare magnesium sulphate concentrations achieved by intramuscular and intravenous regimens used for the prevention of eclampsia.


Low-resource obstetric hospitals in Nagpur and Vellore, India.


Pregnant women at risk for eclampsia due to hypertensive disease.


A pharmacokinetic study was performed as part of a randomised trial that enrolled 300 women comparing intramuscular and intravenous maintenance regimens of magnesium dosing. Data from 258 enrolled women were analysed in the pharmacokinetic study. A single sample was drawn per woman with the expectation of using samples in a pooled data analysis.

Main outcome measures

Pharmacokinetic parameters of magnesium distribution and clearance.


Magnesium clearance was estimated to be 48.1 dl/hour, volume of distribution to be 156 dl and intramuscular bioavailability to be 86.2%. The intramuscular regimen produced higher initial serum concentrations, consistent with a substantially larger loading dose. At steady state, magnesium concentrations in the intramuscular and intravenous groups were comparable. With either regimen, a substantial number of women would be expected to have serum concentrations lower than those generally held to be therapeutic.


Clinical implications were that a larger loading dose for the intravenous regimen should be considered; where feasible, individualised dosing of magnesium sulphate would reduce the variability in serum concentrations and might result in more women with clinically effective magnesium concentrations; and lower dose magnesium suphate regimens should be considered with caution.


Magnesium sulphate (MgSO4) has been demonstrated to treat maternal seizures associated with eclampsia with greater efficacy and fewer complications than either phenytoin or diazepam.[1] It has also been demonstrated to be superior to placebo[2] and nimodipine[3] for the prevention of seizures in women with a diagnosis of pre-eclampsia. Therapeutic serum concentrations of magnesium for the prevention and treatment of seizures have not been rigorously determined. Minimum levels of 4 mEq/l (4.8 mg/dl) have been suggested based on clinical experience rather than a formal evaluation of dose–response.[4-6] Current dosing recommendations are largely based on the regimens used in the MAGPIE Trial.[2] Although the trial confirmed the safety of these empiric regimens and the broad efficacy based on a 50% reduction in seizure rate, the trial did not speak to dose optimisation or the potential benefits of individualised care.

Traditionally, MgSO4 has been administered as MgSO4·7H2O by an intramuscular or an intravenous regimen. Although the total dose of MgSO4 differs between regimens, they have been held to be clinically equivalent.[2] In this study, we examined the exposure to magnesium associated with the two regimens.


The pharmacokinetic (PK) study was conducted as part of a randomised trial comparing two methods of administering MgSO4 to women diagnosed with pre-eclampsia and deemed likely to benefit from treatment with MgSO4 for the prevention of eclampsia. (trial no. NCT00666133). The trial was conducted in two low-resource settings in India: Government Medical College, Nagpur and Christian Medical College, Vellore. The study was approved by the Western Institutional Review Board, the Indian Council on Medical Research (#00000533) and by local institutional review boards at GMC-Nagpur and CMC-Vellore. The primary results of that trial are reported elsewhere including a CONSORT diagram.[7] Women were randomised to receive one of two maintenance regimens (either intravenous infusion or intramuscular) used in the large international Magnesium for the Prevention of Eclampsia (MAGPIE) trial.[2] The MgSO4 was administered as magnesium heptahydrate (MgSO4·7H2O) in a 50% solution. The intravenous group received a loading dose of 4 g (8 ml) intravenously over approximately 20 minutes followed by a maintenance dose of 1 g/hour intravenously. The intravenous infusion was delivered by the Springfusor® pump (Go Medical, Subiaco, Australia). The intramuscular group received a loading dose of 4 g (8 ml) intravenously over 20 minutes by manual push followed by a deep intramuscular loading of 10 g (10 ml in each buttock). One 5 g (10 ml) deep intramuscular dose was then repeated every 4 hours. Over 12 hours, the intravenous group would have received 16 g MgSO4·7H2O and the intramuscular group would have received 24 g, (4 g intravenously and 20 g intramuscularly). The higher total dose of MgSO4 in the intramuscular protocol was originally designed assuming a reduced bioavailability of the intramuscular drug. The goal of the PK study was to assess the pharmacological equivalence of the two regimens, and to provide an estimate of intramuscular bioavailability.

Due to the infrastructure available at the sites, a traditional, more robust PK study involving multiple samples from a smaller number of women was not feasible. A population PK study was designed with a single sample drawn from each woman with the expectation of using samples in a pooled data analysis. The sampling times were chosen to reveal expected PK changes associated with each protocol. The nominal sampling times (relative to the start of the loading intravenous bolus) for the intravenous group were: −01:00, 00:30, 00:45, 03:00, 05:00, 07:00, 11:00, 17:00, 20:00; and for the intramuscular group were: −01:00, 00:30, 04:00, 04:30, 08:00, 10:00, 12:00, 13:00, 16:00, 17:30, 20:00. (times shown in bold type represent pre-intramuscular dosing [trough] samples). These times were chosen to be informative about peak and trough concentrations. The sampling times were assigned to women in a randomised, structured fashion to insure an informative distribution of sampling times. The actual sampling times were used in the analysis.

The analysis of magnesium concentrations was conducted on site in GMC-Nagpur and CMC-Vellore. At GMC-Nagpur, magnesium concentration analysis was performed with the Selectra E (Merck Chemicals, Mumbai, India). At CMC-Vellore magnesium concentration was measured with Autopure Magnesium (Hitachi, Roche Diagnostics India, Mumbai, India).

The PK modelling of magnesium concentration was conducted as a mixed effects, population analysis of PK data (commonly termed a ‘population PK analysis’). The population PK analysis characterises the central tendency as well as multilevel variability (between-subject variability and residual unknown variability) in concentration–time profile data. In this analysis, because there was only a single data point per woman, we did not attempt to model the between-subject variability. The population PK analysis allowed for subsequent investigation of subject-specific co-variates[8, 9] such as serum creatinine concentration and maternal body weight to explain some of the residual unknown variability estimated by the model.

One- and two-compartment linear elimination models have been used previously to describe the PK of magnesium.[10, 11] The simpler one-compartment model, with one additional compartment for first-order intramuscular absorption, was chosen here to describe the data, again because of availability of a single data point per woman. The maximum likelihood population parameter estimates were determined for clearance (CL), volume of distribution (V), intramuscular absorption rate (KA), intramuscular bioavailability (F) and baseline endogenous steady-state magnesium concentration (BL). The administered magnesium was modelled as additive to BL. Additive (constant standard deviation) and proportional (constant fractional standard deviation) residual error models were both tested for goodness of fit. Estimation of these parameters constituted the ‘Base Model’. Analysis was performed using NONMEM 7.[12]

After determination of the Base Model, creatinine concentration and body weight were tested as model covariates for potential inclusion in the Final Model based on plausible mechanistic assumptions. Magnesium is cleared by renal filtration, so estimated clearance should be partially predicted by serum creatinine concentration. Maternal body weight should reflect volume of distribution (V). We considered a relationship between V and body weight (WT) of the form: Vj = V × (WTj/55)θ, where Vj is the volume of distribution for subject j, V is the typical volume of distribution for a subject of WT 55 kg (the median study subject weight) and θ is the weight effect exponent. Creatinine clearance as a measure of glomerular filtration rate was not available, so an inverse relationship between CL and serum creatinine concentration was tested. The Final Model inclusion of a covariate relationship was based on statistical significance.[13]


In the randomised trial, 147 women were enrolled in the intravenous arm and 153 in the intramuscular arm. Thirty-five of these women were excluded because no magnesium was sample-drawn or treatment was interrupted before magnesium sampling. Seven additional women were excluded because sampling times were determined to have been mislabelled as time zero rather than at the first anticipated peak level. The clinical characteristics of the women in this analysis are described in Table 1. They are not substantively different from those in the primary study.

Table 1. Clinical characteristics of the cohort
  Intravenous ( = 126) Intramuscular ( n  = 132)
Age (years)
Mean (SD)25 (4.3)24 (3.6)
Weight (kg)
Mean (SD)57.3 (11.3)55.6 (11.3)
Gestational age (weeks)
Mean (SD)33.7 (3.7)34.1 (4.2)
Systolic blood pressure (mmHg)
Mean (SD)154 (14.5)154 (13.0)
Diastolic blood pressure (mmHg)
Mean (SD)103 (9.3)103 (8.4)
Trace (%)1 (0.8)0 (0)
+1 (%)38 (30.2)38 (28.8)
+2 (%)42 (33.3)42 (31.8)
+3 (%)31 (24.6)39 (29.5)
+4 (%)14 (11.1)13 (9.8)

A one-compartment model with proportional variability fitted the data well, especially considering the necessity for a pooled analysis approach to fit data with such large inherent variability. The model parameter precisions were generally good (<16% SE) except for intramuscular absorption rate KA (34% SE) and covariate effect exponents (<25% SE).

The Base Model concentration–time curves are compared by group data in Figure 1 with box-and-whisker plots, where the median data are marked by a black bar and the box represents the 25th to 75th centiles of data. All time zero data points were grouped into a single box-and-whisker plot, plotted at time 0. In the third panel, the concentration–time curves for intramuscular and intravenous administration are presented with scatter plots of individual measurements.

Figure 1.

(A) Base PK model compared with the data for intramuscular dose. (B) Base PK model compared with the data for intravenous dose. The data are represented as box-and-whisker plots. (C) Base model predicted and observed concentration values for intramuscular and intravenous modes are superimposed with a scatter plot of data points for ease of comparison.

In the Final Model, volume of distribution (V) was found to depend on weight (WT), and clearance (CL) was found to depend inversely on serum creatinine concentration. The inclusion of each covariate in the model (as predictors of V and CL) improved the model fit and was significant at < 0.001. CL was estimated to be 48.1 dl/hour; V was estimated to be 156 dl and intramuscular bioavailability was estimated to be 86.2%. All Final Model parameter estimates and precisions are presented in Table 2. The inclusion of individual measures of serum creatinine and weight as predictors of V and CL, respectively, in the Final Model reduced the residual coefficient of variability (CV) to 22.9% from 25%, which was significant at < 0.001.

Table 2. Final PK model parameter estimates and%SE (standard error)
   Value SE
  1. Vj = V*(WTj/55)θ1 where Vj is the volume of distribution for subject j.

  2. CLj = CL*(0.8/CrConcj)θ2 where CLj is the clearance for subject j.

  3. WTj and CrConcj are body weight and creatinine concentration for subject j.

  4. a

    Not appropriate for intravenous administration.

Volume of distributiondl1568%
Intramuscular absorption ratea, KA/hr0.31732%
Intramuscular bioavailabilitya, F 0.86216%
Baseline magnesium concentrationmmol/l0.853.2%
Weight effect exponent (θ1) 0.69227%
Serum creatinine effect exponent (θ2) 1.4818%
Residual variability%CV22.911%

Figure 2 shows observed versus predicted magnesium concentrations for the Base Model and Final Model, respectively. These diagnostic plots show that the inclusion of covariates substantially improves prediction of magnesium concentration. The model and data are in generally good agreement, but further clinical variability in magnesium concentration is evident.

Figure 2.

Observed versus predicted magnesium concentration for Base and Final PK models.

Figure 3 shows simulations based on pharmacokinetic parameters derived for hypothetical women of various body weights and serum creatinine concentrations for the each group respectively. These plots were designed to provide information about the range of possible time–courses observed in this population. Each figure shows simulated concentration–time profiles for nine simulated ‘typical’ women, representatives of combinations of the 5th, 50th or 95th centiles of body weights: 42, 55 or 78 kg and the 5th, 50th or 95th centiles of study creatinine concentrations: 45.8, 61.0 or 91.5 μmol/l, (0.6, 0.8 or 1.2 mg/dl). The differences in body weight result in substantial differences in concentration immediately after the bolus, presumably as a result of differences in volume of distribution. At 12 hours, the differences in magnesium concentration are more closely associated with the differences in serum creatinine concentration, presumably as the result of the differences in renal filtration.

Figure 3.

Concentration-time profiles for nine simulated ‘typical’ subjects spanning the range of covariate variability (5th, 50th and 95th centile of body weight and of creatinine concentration) for intramuscular and intravenous dose groups.

The typical concentration–time profiles for the intravenous and intramuscular dose groups are compared with simulated profiles for intravenous dosing where the loading dose was increased from 4 g to 5, 6, 7 and 8 g, respectively (Figure 4). As would be expected, the results indicate that an increased loading dose could provide a higher concentration–time profile in the first 6–8 hours, more comparable to the intramuscular concentration–time profile. The differences become vanishingly small by 14 hours post dose initiation as women approach steady state.

Figure 4.

Typical concentration–time profiles for the intravenous and intramuscular dose groups, superimposed with simulations of intravenous dosing with loading dose increased from 4 g to 5, 6, 7 and 8 g.


Main findings

Magnesium clearance was found to be 48.1 dl/hour, (80.1 ml/minute), comparable to the glomerular filtration rate that might be expected from a cohort of women with pre-eclampsia and similar to the 50 dl/hour found by Lu et al.[11] and 42.8 dl/hour by Chuan et al.[10] The volume of distribution was 156 dl, 30–50% lower than previously reported. This discrepancy may be, in part, attributed to the small size of the women in this study. The bioavailability of intramuscular magnesium was found to be 86.2%. The serum magnesium concentrations immediately after the loading dose were significantly lower in the intravenous arm. Both regimens produced broad variability in concentration–time profiles across the population studied. (Figure 1).

Strengths and weaknesses

The randomised trial provided a unique opportunity to evaluate serum magnesium concentrations over a large number of women, permitting us to describe a wide range of variability and to explore determinants of that observed variability. Our study has a number of limitations. The PK analysis was performed with a single sample from each woman. Therefore, the variability between women could not be estimated separately from the residual variability. However, treating this number of women with an investigational intramuscular regimen in a developed world setting, where a more robust PK study could be performed, would be difficult. Second, the population studied is made up of women from India whose body mass is typically substantially less than that of Western women and would therefore be expected to have smaller volumes of distribution. Accordingly, our results suggest initial serum concentrations higher than might be expected in a population of larger women, reinforcing the consideration for a larger loading dose with the intravenous regimen.


The variability in serum magnesium concentration was surprisingly large for an ion not subject to metabolism. Whereas the variability was not present at baseline before treatment, it was immediately observed after the intravenous bolus but before the potential impact of differences in renal clearance. The differences in between-subject volume of distribution may partially explain this variability. We found that maternal weight accounted for a portion of the observed variability. We hypothesise that women with pre-eclampsia have considerable between-subject differences in intravascular space and in oedema impacting their extravascular space, each impacting the volume of distribution of magnesium. Serum creatinine, as a reflection of glomerular filtration rate, was found, as expected, to have an inverse relationship to clearance and would also be expected to contribute to observed variability, as we found. Women with more severe pre-eclampsia would be expected to have higher serum creatinine concentrations and therefore higher magnesium levels at steady state. The anticipated concentration–time profiles of women with various weights and serum creatinine concentrations are described in Figure 3.

The differences in concentration–time curves from intramuscular and intravenous dosing regimens could be clinically significant (Figure 1). Immediately after loading doses, the serum magnesium concentrations for women in the intramuscular arm are considerably higher than for the intravenous arm, consistent with the additional 10 g of MgSO4·7H2O loading dose given intramuscularly. Given a higher likelihood of seizures early in the treatment of pre-eclampsia, higher serum concentrations earlier in the course of therapy, consistent with those in the intramuscular arm, may be desirable. Figure 4 shows the potential impact of higher intravenous loading doses in the intravenous arm. Loading doses of 6–7 g would achieve concentrations comparable to those in the intramuscular arm.

At 12–14 hours, the apparent steady-state concentrations in each group were similar. The intravenous group received a maintenance dose of 4 g every 4 hours; 80% of the intramuscular maintenance dose of 5 g every 4 hours. Given a bioavailability of 86% for intramuscular magnesium, comparable steady state concentrations at this dosing regimen would be expected.

In a small PK study comparing the intramuscular and intravenous regimens used in our protocol, Sibai et al.[14] also reported lower magnesium concentrations in the intravenous group compared with the intramuscular group early in the course of therapy, similar to our results. In a study using goats to compare the PK of magnesium, Hankins et al.[15] also reported significantly lower magnesium concentrations in the first 4 hours of therapy for intravenous versus intramuscular regimens.

Therapeutic serum concentrations of magnesium for the prevention and treatment of eclampsia have not been rigorously determined. Minimum levels of 4 mEq/l (4.8 mg/dl) have been suggested.[4-6] Approximately 25% of women in the intravenous arm of the model, represented by the lower limit of the box in Figure 1B, had serum magnesium concentrations <2.9 mEq/l, (3.5 mg/dl). Similarly, approximately 25% of women in the intramuscular arm had serum trough concentrations <3.3 mEq/l, (4.0 mg/dl) (Figure 1A). The magnesium dosing used in this study was the same as those used in the MAGPIE Trial, where the seizure rate was reduced by 50%–0.8% of treated women. The seizure rate in each arm of our trial, 0.7%, was comparable to the MAGPIE Trial. Given the low concentrations detected in some women, a higher dose of MgSO4 might be expected to result in a lower seizure rate more consistent with those reported in regions such as the USA. A higher dose would, however, result in higher concentrations and a greater potential for toxicity. In low-resource settings where serum magnesium concentrations cannot be routinely monitored and the infrastructure to diagnose and respond to magnesium toxicity is less robust, higher dosing could be problematic. The low incidence of eclampsia in this study limits our ability to recommend optimal magnesium dosing. Our results do confirm significant variability among individuals that could be clinically relevant. When possible, the individualisation of dosing based on maternal weight and serum creatinine concentration followed up with monitoring of serum magnesium concentration and dose adjustment based on results may be needed to achieve the greatest improvement in seizure prevention.


In summary, we have performed model-based PK analysis on the concentration–time data from women with pre-eclampsia treated with MgSO4. Several important clinical observations can be made. First, the 4-g loading dose routinely used in intravenous regimens provides lower initial concentration than achieved with the intramuscular regimen. Our modelling suggests that an increased intravenous loading dose (6 g) would produce initial concentrations similar to that observed with the intramuscular regimen. Second, serum concentrations are low, and possibly subtherapeutic, in a significant percentage of women in both groups. Some of the variability is due to differences in maternal weight, as a reflection of volume of distribution, and to differences in serum creatinine concentration, as a reflection of glomerular filtration rate. The differences in extravascular volume due to oedema associated with pre-eclampsia may also contribute to the variability in volume of distribution. When local resources permit, individualisation of dosing could be considered. Finally, some have suggested use of an MgSO4 regimen with a lower dose than that used in the MAGPIE Trial.[16] Our data suggest that a lower dose would result in a substantial number of women with low magnesium concentrations. As the optimal therapeutic concentration of magnesium is unknown, it is uncertain if this might be reflected in a higher rate of eclampsia.

Disclosure of interests


Contribution of authorship

DHS contributed to design of the PK sampling protocol, primary modeling analysis and manuscript preparation. SM contributed to design of the primary randomised trial, direct study coordination at GMC-Nagpur and manuscript review. AR contributed to the direct study coordination at CMC-Vellore and manuscript review. HB contributed to the design of the primary randomised trial, study coordination at GMC-Nagpur and CMC-Vellore, research site visits, data quality coordination and manuscript review. BW contributed to the design of the primary randomised trial, grant coordination and management and to manuscript review. PV performed the senior modeling analysis and contributed to manuscript review. TE contributed to the design of the primary randomised trial and the PK sampling protocol, to study coordination at GMC-Nagpur and CMC-Vellore, research site visits and manuscript preparation.

Details of ethics approval

The study received ethics approval from the Western institutional review board on 29 January 2008 (#00000533) and from Government Medical College—Nagpur, Christian Medical College—Vellore and the Indian Council on Medical Research.


Financial support was received from the Macarthur Foundation.


Go Medical, Subaico, Australia provided the Springfusor pump and flow control tubing at a reduced rate but did not contribute to the design or analysis of the study.