ARTICLES

Effects of redesigned community postnatal care on womens health 4 months after birth: a cluster randomised controlled trial
C MacArthur, H R Winter, D E Bick, H Knowles, R Lilford, C Henderson, R J Lancashire, D A Braunholtz, H Gee

Summary
Background Much postpartum physical and psychological morbidity is not addressed by present care, which tends to focus on routine examinations. We undertook a cluster randomised controlled trial to assess community postnatal care that has been redesigned to identify and manage individual needs. Methods We randomly allocated 36 general practice clusters from the West Midlands health region of the UK to intervention (n=17) or control (19) care. Midwives from the practices recruited women and provided care. 1087 (53%) of 2064 women were in practices randomly assigned to the intervention group, with 977 (47%) women in practices assigned to the control group. Care was led by midwives, with no routine contact with general practitioners, and was extended to 3 months. Midwives used symptom checklists and the Edinburgh postnatal depression scale (EPDS) to identify health needs and guidelines for the management of these needs. Primary outcomes at 4 months were obtained by postal questionnaire and included the womens short form 36 physical (PCS) and mental (MCS) component summary scores and the EPDS. Secondary outcomes were womens views about care. Multilevel analysis accounted for possible cluster effects. Findings 801 (77%) of 1087 women in the intervention group and 702 (76%) of 977 controls responded at 4 months. Womens mental health measures were significantly better in the intervention group (MCS, 303 [95% CI 153452]; EPDS 192 [255 to 129]; EPDS 13+ odds ratio 057 [043076]) than in controls, but the physical health score did not differ. Interpretation Redesign of care so that it is midwife-led, flexible, and tailored to needs, could help to improve womens mental health and reduce probable depression at 4 months postpartum. Lancet 2002; 359: 37885 See Commentary page 370

Introduction
Results of studies14 have shown that many women have, but do not report, physical and emotional disorders after childbirth, some of which are persistent. At present, postnatal care does not adequately address these needs, but continues to devote much time and resources to routine examinations that screen for morbidities that are no longer the major health burden for women.58 UK Government reports and other national bodies have clearly stated the need for wide-ranging changes to maternity services, emphasising poor assessment and frequently inappropriate delivery of postnatal care.912 A service led by midwives, with continuity of care and involvement of women, which is supportive and sensitive to individual needs and preferences is at the centre of the maternity care recommendations.10 We aimed to develop and implement a new model of community postnatal care, that was based on results of research and maternity care recommendations, and to compare the effects of such care on womens subsequent health with those of present care.

Patients and methods

Recruitment and randomisation The unit of randomisation was the general practice. Since the intervention included midwife education, and midwives could not be expected to deliver two models of care without contamination, randomisation of individual women was not possible. Midwives could also not be randomly assigned because of contamination from crossover in care by midwives in the same practice. General practices were randomly selected from all those in the West Midlands health region. Women were eligible if they had had postnatal care in the recruited practices between October, 1997, and April, 1999. Women were informed about the study between 34 weeks gestation and the first home visit. Written informed consent was obtained. Only women who expected to move out of the general practice in the postnatal period were excluded. Estimates of the number of eligible women per cluster were obtained from health authorities and from practices. All 17 ethics committees in the region gave approval. We randomly selected 125 general practitioners from those registered in the West Midlands health region. Agreement from all partners in a practice and from midwifery managers in the local UK National Health Service organisation was sought before the midwives were approached for participation in the trial. Consenting cluster practices were randomly assigned at the training days by a member of the Birmingham Clinical Trials Unit who was independent of the trial team, with a customised computer program. Minimisation was used, which gives the next cluster to enter the trial a probability of greater than 05, whichever treatment would keep to a minimum the overall imbalance between groups at that stage of the trial13 in relation to Townsend score and estimated maternity caseload of the practice. Masking of health professionals or participants was not possible.

Department of Public Health and Epidemiology, University of Birmingham, Edgbaston, Birmingham, UK (Prof C MacArthur PhD, H R Winter MD, Prof R Lilford MFPHM, R J Lancashire BA, D A Braunholtz BSc); RCN Institute, Radcliffe Infirmary, Oxford (D E Bick MMedSc); Institute of Health, Bournemouth University, UK (H Knowles MPhil); School of Health Sciences/Nursing, University of Birmingham, Edgbaston, Birmingham (C Henderson MA); Birmingham Womens Hospital NHS Trust, Edgbaston, Birmingham (H Gee FRCOG) Correspondence to: Prof Christine MacArthur (e-mail: C.MacArthur@bham.ac.uk)

378

THE LANCET Vol 359 February 2, 2002 www.thelancet.com

For personal use. Only reproduce with permission from The Lancet publishing Group.

ARTICLES

Intervention Our new model of community-based postnatal care meant that care could be tailored flexibly to individual needs. Care was led by midwives, with contact with general practitioners based on referral, including home visits and the final discharge consultation. To ensure that specific needs could be identified, even if not spontaneously reported by the women or observed by the midwife, a symptom checklist was used at the first visit (immediate symptoms only), at days 10 and 28, and at the discharge consultation at 1012 weeks. The Edinburgh postnatal depression scale (EPDS) was also used to screen for depression at day 28 and at the discharge consultation. Care plans were made and visits scheduled on the basis of these results so that care could be tailored to individual needs rather than based on a predetermined schedule. We developed ten evidence-based guidelines for the main postpartum disorders identified from published work for subsequent midwife management of physical and psychological disorders, all of which had clear criteria for referral to general practitioners. Every guideline was peer-reviewed by national experts,14 and summarised in a leaflet. A description of how to use the checklists and guidelines to make visits more flexible was also included. To fully appraise womens health over a longer period than usual, we extended care so that the last home visit was routinely at 28 days and women had their discharge consultation at 1012 weeks. Since general practitioners were not required to have routine postnatal contact, advice usually given by them, such as the need for contraception and immunisation, was given by the midwives, with referral to their general practice or relevant clinic as needed. Present community postnatal care in the UK generally consists of about seven midwife home visits to 1014 days (can continue to day 28) after birth, and care from health visitors thereafter. Because some health visitors use the EPDS to screen for depression, some controls would also have been screened with this test. General practitioners did routine home visits and a final 68 week check. Training Equal attention was paid to midwives in both the intervention and control groups to keep the possibility of a differential Hawthorne effect to a minimum.15 For initial training, all midwives attended a study day with the team, which was approved by the English National Board. We trained all midwives in postnatal care and health and trial design. Midwives in the intervention group were then trained to implement the new model of care, while those of the control group discussed studies of midwifery practice, with and without adequate control, to emphasise the role and importance of being controls. We gave brief written information as appropriate to each group. Midwives in the intervention group were given a copy of the guidelines, emphasising that these must not be shown to any other midwife. Midwives who joined trial practices during the study period received training locally. Continuing contact for midwives in both groups consisted of a monthly visit from a team midwife, daily telephone availability, and monthly newsletters. Data collection Baseline data for patients were obtained from three sources. The consent form included date of birth and postcode address, from which a Townsend deprivation score16 was derived. At the first home visit, midwives gave

women a brief questionnaire to complete and return to the study office. This questionnaire recorded household composition, age of leaving full-time education, home ownership, and assessment of social support. The social support score was based on womens responses to five items: a visual analogue scale rating the amount of help from her partner; availability of a friend or relative to contact if needed; a local contact who has a baby; availability of practical help at home; and duration of residence in the locality. Data for parity, mode of delivery, and perineal trauma were obtained from the obstetric units, or from general practitioner records for home births. Cluster baseline data included Townsend score for practice postcode, cluster size, number of partners in the practice, number of midwives, and midwife qualifications. Since most clusters had more than one midwife, the cluster average for the highest qualification per midwife was used and stratified into three levels. The number and duration of midwife visits were recorded to describe what actually happened and to form a basis for calculation of resource consequences. In the intervention group, these measures were recorded by the midwife as part of the customised care record, whereas midwives in the control group continued to complete their usual Trust records, and kept a separate diary to note only the date and duration of each visit to women in the trial. Since this diary was an additional record with care otherwise unchanged, we expected under-recording in the control group. Information from the diaries was therefore supplemented by data from a 28-day diary that was given to women in both groups to note all home visits made by midwives, general practitioners, and other health professionals. Data on outcomes were obtained by postal questionnaire at 4 months, with a reminder 3 weeks later. Questionnaires were returned to the study office in prepaid envelopes. The number of home visits by general practitioners was also recorded in the 4-month questionnaires. Outcome measures Primary outcome measures used to assess the womens health and wellbeing at 4 months were the summary physical and mental component scores (PCS and MCS) of the short form 36 (SF36) general health questionnaire17 and the EPDS.18 Secondary outcomes were womens views about care. Overall satisfaction with care from community midwives was scored as very satisfied or satisfied, compared with very dissatisfied, dissatisfied, or neither satisfied nor dissatisfied. Care relative to expectations was scored as much better or better, compared with much worse, worse, or the same. A planning care score was derived from womens views of the appropriateness of the number of midwife home visits, extent of involvement in planning care, and overall view on whether care was planned. Continuity of care score was based on the number of different midwives making visits, how many times the woman knew the name of the next midwife to visit, and how many times she felt that she received conflicting advice. A maternity discharge consultation score was based on the womans general satisfaction with this visit, the extent to which health problems were discussed, and whether they felt the timing was right, too early, or too late. Ability to discuss health symptoms was assessed by whether women felt able to talk to midwives about all or most, compared with a few, some, or no symptoms. Difficulty in talking about symptoms was scored as any difficulties or none.

THE LANCET Vol 359 February 2, 2002 www.thelancet.com

379

For personal use. Only reproduce with permission from The Lancet publishing Group.

ARTICLES

120 practices approached 37 practices excluded 31 midwife manager restrictions 2 midwives not attached to practice 2 health visitor objections 2 individual GPs not traced 83 practices eligible

43 19 23 1

practices did not take part declined failed to respond midwife declined

40 practices took part but in three cases midwives gave care in a second practice

37 clusters randomised

18 clusters randomised to intervention 1 practice midwife went on long-term sick leave 17 clusters recruited women 1830 women eligible 743 not recruited 1087 women recruited 24 withdrew 22 moved 2 babies died 1027 obstetric unit data available 992 returned demographic questionnaire 783 returned women's diary 958 midwife diary available

19 clusters randomised to control

19 clusters recruited women 1750 women eligible 773 not recruited 977 women recruited 13 withdrew 12 moved 1 baby died 887 obstetric unit data available 858 returned demographic questionnaire 542 returned women's diary 759 midwife diary available

801 women returned 4-month outcome questionnaire

702 women returned 4-month outcome questionnaire

Figure 1: Trial profile

Systematic recording of reasons for non-recruitment was prohibitive because of midwife time, but monthly meetings recorded general summaries of reasons, most of which were organisational: refusal by women was rare.

380

THE LANCET Vol 359 February 2, 2002 www.thelancet.com

For personal use. Only reproduce with permission from The Lancet publishing Group.

ARTICLES

Statistical analysis In estimating sample size, we aimed to accurately represent sociodemographic characteristics and primary care settings. Using modelling of these attributes, we estimated that about 40 general practices would be needed. Original estimates, which were for a threegroup trial, suggested that the study would have 80% power to detect a two-point difference in the physical components score (PCS) of the SF36 (type 1 error at 5%). After amending these power calculations to account for a two-group trial, and with additional information from practices and midwives at practice recruitment, estimation of sample size was checked. These calculations suggested that around 4000 women in 40 practices would be needed. With a 25% loss to recruitment and follow-up, outcome measures would be available for 3000 women. This sample size would give an 80% power to detect a two-point difference between the groups in the PCS. Such a difference is smaller than that judged likely to have a meaningful clinical difference. This calculation is based on an SD of 1042 for the PCS from data for 1412 women of all ages in the USA,17 and an intraclass correlation of 001 (reported for a quality-of-life score for asthma on assessment of area-wide and organisational interventions).19 The intraclass correlation is a measure of the proportion of variation in outcome measures that can be attributed to possible non-independence of women in each cluster. We used SPSS version 10 for univariate analyses. In analysing cluster randomised data, it was necessary to allow for the possibility that outcome measures from individuals within a cluster might not be independent. For example, factors predisposing women to adverse psychological wellbeing could differ by the general practice catchment. We used random effect multilevel modelling, in which the hierarchical nature of the data is recognised, and variations attributable to individual and to general practice are estimated. In the multilevel models, using MLwiN version 1.120 we included two levels of variation, individual woman and general practice. Odds ratios and 95% CIs were calculated for categorical outcomes and coefficients for continuous measures. All group comparisons were prespecified and analysis was by intention-to-treat.

Role of the funding source The sponsors of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report.

Results
From a randomly selected list of 125 general practitioners, we identified 120 practices to approach (five were selected twice through two partners), from which 40 (33%) took part. In three cases, midwives also gave care in another recruited practice, but each practice pair was included as one cluster. We thus had 37 general practice clusters which were allocated either to intervention (n=18) or to control (19). One practice was excluded after allocation but before recruitment of women, because the sole midwife in this single general practitioner practice went on long-term sick leave. 17 practices in the intervention group and 19 in the control
Intervention (n=1087) Parity 0 1 2 Missing Maternal age 19 2024 2529 3034 35 Missing Mode of delivery SVD Instrumental Section Missing Perineal trauma 1st degree tear 2nd degree tear 3rd/4th degree tear Episiotomy Intact Missing 448 (41%) 364 (33%) 249 (23%) 26 (2%) 61 (6%) 184 (17%) 320 (29%) 337 (31%) 172 (16%) 13 (1%) 693 (64%) 116 (11%) 227 (21%) 51 (5%) 100 (12%) 156 (19%) 4 (05%) 175 (22%) 344 (43%) 30 (4%) Control (n=977) 435 (45%) 324 (33%) 171 (18%) 47 (5%) 72 (7%) 158 (16%) 329 (34%) 285 (29%) 122 (12%) 11 (1%) 635 (65%) 107 (11%) 151 (15%) 84 (9%) 114 (15%) 143 (19%) 1 (01%) 154 (21%) 287 (39%) 43 (6%) 93 (10%) 665 (68%) 97 (10%) 122 (12%) Non-responders (n=561)* 215 (38%) 162 (29%) 131 (23%) 53 (9%) 67 (12%) 133 (24%) 154 (27%) 129 (23%) 56 (10%) 22 (4%) 353 (63%) 49 (9%) 96 (17%) 63 (11%) 59 (15%) 60 (15%) 3 (08%) 58 (14%) 203 (50%) 19 (5%) 80 (14%) 269 (48%) 54 (10%) 158 (28%) 305 (54%) 98 (17%) 158 (28%) 158 (28%) 101 (18%) 123 (22%) 179 (32%) 197 (35%) 201 (36%) 163 (29%) 94 (17%) 106 (19%) 140 (25%) 203 (36%) 18 (3%)

Intervention (n=17)*

Control (n=19)

Nonrecruited practices (n=83)

Other adults in house 0 155 (14%) 1 725 (67%) 2 106 (10%) Missing 101 (9%)

Cluster GP partners 1 1 (6%) 25 6 (35%) 6 10 (59%) Midwives 1 3 (18%) 2 14 (82%) Townsend score (mean [SD]) 149 (39) Midwife qualification score Lower 6 (35%) Medium 5 (30%) Higher 6 (35%) Cluster size (women recruited) 150 11 (65%) 51100 2 (12%) 101 4 (23%)

Age at completion of full-time education 18 686 (63%) 653 (67%) 19 302 (28%) 203 (21%) Missing 99 (9%) 121 (12%) Social support score 12 1314 15 Missing Home ownership Owned Rented Missing Townsend quartiles Most affluent Affluent Deprived Most deprived Missing 359 (33%) 273 (25%) 323 (30%) 132 (12%) 621 (57%) 357 (33%) 109 (10%) 289 (27%) 235 (22%) 237 (22%) 297 (27%) 29 (3%) 307 (31%) 206 (21%) 314 (32%) 150 (15%) 638 (65%) 214 (22%) 125 (13%) 217 (22%) 258 (26%) 266 (27%) 206 (21%) 30 (3%)

4 (21%) 9 (47%) 6 (32%) 6 (32%) 13 (68%) 132 (33) 12 (63%) 4 (21%) 3 (16%) 11 (58%) 6 (32%) 2 (10%)

18 (22%) 40 (48%) 25 (30%)

120 (33)

GP=General practioner. *The other empty cluster had one GP, one midwife and lower midwife qualification score. Mean Townsend score, including this cluster, was 137.

Values are number (%). *The 37 women who were withdrawn and 60 who did not receive a 4-month questionnaire are included in non-responders. Vaginal births only (n=1551).

Table 1: Characteristics of clusters by study group

Table 2: Characteristics of women by study group

THE LANCET Vol 359 February 2, 2002 www.thelancet.com

381

For personal use. Only reproduce with permission from The Lancet publishing Group.

ARTICLES

PCS Mean of cluster means* Control Intervention Difference (95% CI) p-value 4784 4668 117 (252 to 019) 0089

MCS 4754 5050 296 (116 to 477) 0002

EPDS 806 640 166 (249 to 083) <00001

EPDS 13 2125% 1439% 685% (1199 to 171) 0010

Multilevel model Study group Control Reference Intervention (odds ratio 95% CI) 079 (191 to 034) 2 (1 df) 189 p value 01692 Mean reference values 4757 Intra-class correlation 000002

Reference 303 (153 to 452) 1577 <00001 4774 000534

Reference 192 (255 to 129) 353 <00001 817 000404

Reference 057 (043 to 076) 1514 00001

For physical component score (PCS) and mental component score (MCS) a higher score is better and for Edinburgh postnatal depression scale (EPDS) a lower score is better. *Unweighted mean of the cluster means. Values are coefficients except where indicated.

Table 3: Physical and psychological health measures at 4 months

group thus recruited 1087 (53%) and 977 (47%) women, respectively. Figure 1 shows the trial profile. Care was given by 42 and 38 midwives, respectively, including all midwives providing care during the study, some of whom moved before recruitment ended and were replaced. The total number of midwives in the intervention and control groups, per cluster ranged from one to five, with a maximum of three at any one time. None of the characteristics of the cluster differed significantly between study groups (table 1). Recruitment rates of women did not differ significantly between the groups (p=014). Table 2 shows baseline characteristics of women by study group. Multilevel model analyses to test whether the characteristics differed more than would be expected given cluster randomisation, showed no significant differences. There were some proportionate differences however, and for more of the characteristics generally indicative of worse health outcome, the differences were biased against the intervention group. The response rate at 4 months was 77% from the intervention group and 76% from controls. Obstetric characteristics did not vary but responders were significantly more likely to be older (p<00001), from less deprived regions (p<00001), to live with another adult (p<00001), and be owner-occupiers (p<00001) (table 2). This pattern of response did not differ between groups. The mental health score (MCS) was significantly better in the intervention group than controls, as were mean EPDS and the proportion of women with an EPDS score of 13 or more, which is indicative of a higher risk of being depressed (table 3). The physical health score (PCS) did not differ between study groups. Table 3 also shows the study group cluster means for the measures and figure 2 shows the distribution of individual cluster means. The associations with study group were confirmed in further multilevel models to adjust for possible residual confounders (characteristics shown in tables 1 and 2) and the effect sizes for MCS and EPDS increased a little (table 4). Overall satisfaction with care from the community midwives did not differ between groups but women in the intervention group were significantly more likely than controls to rate care as better than expected. The planning care score was significantly higher in the intervention group than in controls, but the continuity of care and the maternity discharge consultation scores did not differ. Being able to talk to midwives about most or all symptoms, and having no difficulties in this, were both reported significantly more by women in the intervention group than by controls (table 5).

Data recorded by midwives on the frequency of midwife home visits differed from that recorded by women (table 6). The mean number of visits recorded by midwives was significantly greater in the intervention group than in controls. The mean number of midwife visits recorded in the womens 28-day diaries however, was much the same in both groups, and close to the number recorded by the intervention midwives. The duration of home visits was recorded only in the midwife
A 80 Mean MCS score 70 60 50 40 30 20 10 0 B 80 Mean PCS score 70 60 50 40 30 20 10 0 C 30 Mean EPDS score 20 10 0 Intervention Control
Cluster Study mean

10

Figure 2: Mean cluster scores of mental component score (A), physical component score (B), and Edinburgh postnatal depression scale (C)
Bars are 95% CIs.

382

THE LANCET Vol 359 February 2, 2002 www.thelancet.com

For personal use. Only reproduce with permission from The Lancet publishing Group.

ARTICLES

PCS Study group Control Intervention Maternal age-group 2024 years 19 years 2529 years 3034 years 35 years Parity 0 1 2 Other adults in house 1 0 2 Mode of delivery Normal Instrumental Caesarean section Townsend quartiles Quartile 1 (most affluent) Quartile 2 (affluent) Quartile 3 (deprived) Quartile 4 (most deprived) Social support score Most support Medium support Least support Cluster size 101 51100 50 Mean reference values

MCS

EPDS

EPDS 13 or more (odds ratio [95% CI]) Reference 047 (031 to 076) Reference 202 (089 to 459) 099 (057 to 170) 128 (073 to 223) 188 (102 to 347) Reference 129 (088 to 190) 169 (105 to 270) Reference 185 (116 to 293) 131 (079 to 219) Reference 156 (085 to 287) 099 (063 to 154) Reference 118 (075 to 185) 127 (079 to 204) 166 (098 to 282) Reference 145 (095 to 220) 202 (140 to 293) Reference 091 (055 to 150) 113 (068 to 188)

Reference 080 (232 to 072) Reference 360 (713 to 008) 136 (062 to 334) 010 (197 to 218) 007 (246 to 233) Reference 061 (203 to 079) 121 (306 to 064) Reference 234 (434 to 035) 196 (402 to 010) Reference 075 (307 to 158) 402 (576 to 228) Reference 087 (076 to 249) 134 (309 to 040) 083 (123 to 288) Reference 067 (218 to 084) 161 (301 to 022) Reference 036 (136 to 208) 220 (046 to 395) 4964

Reference 431 (250 to 612) Reference 448 (867 to 028) 042 (278 to 195) 108 (355 to 139) 268 (553 to 018) Reference 129 (297 to 039) 490 (711 to 270) Reference 122 (360 to 116) 129 (117 to 374) Reference 278 (555 to 001) 126 (334 to 081) Reference 108 (302 to 086) 154 (362 to 054) 284 (528 to 039 Reference 104 (284 to 076) 311 (477 to 145) Reference 080 (285 to 125) 217 (425 to 009) 5247

Reference 268 (346 to 189) Reference 185 (003 to 367) 038 (140 to 064) 004 (111 to 103) 090 (033 to 214) Reference 083 (010 to 156) 178 (083 to 273) Reference 113 (010 to 216) 001 (107 to 104) Reference 059 (060 to 178) 075 (015 to 165) Reference 040 (045 to 124) 087 (003 to 177) 105 (002 to 211) Reference 066 (012 to 143) 153 (081 to 225) Reference 005 (095 to 084) 044 (046 to 133) 626

Values are coefficients (95% CI) except where indicated. PCS=physical component score. MCS=mental component score. EPDS=Edinburgh postnatal depression scale.

Table 4: Multilevel model of physical and psychological health measures by study group, including adjustment for other characteristics

records and was greater in the intervention group than in controls for the first visit (413 min [SD 114] and 306 min [62], p<00001) and for subsequent visits (302 min [SD 72] and 253 min [52], p<00001). On the basis of the number of midwife-recorded visits, the mean total midwife visit duration was 1923 min (SD 354) in the intervention and 1083 min (346) in controls; and based on the number of visits recorded by the women, mean duration was 1869 [SD 330] and 1568 [SD 258], respectively. The mean number of home visits by a
Intervention/ control comparison Views Overall satisfaction* Better than expected* Planning care score (max 18, median 16) Continuity of care score (max 14, median 12) Maternity discharge score (max 11, median 9) Talk to midwife about most/all health symptoms* No difficulty in talking to midwife about health symptoms* 109 (072 to 163) 135 (108 to 170) 049 (013 to 085) 021 (011 to 052) 019 (011 to 050) 152 (105 to 220) 161 (107 to 241) 2 (1 df) p

general practitioner recorded by the women in both their diaries and the 4-month questionnaires was similar (table 6). A record of the postnatal discharge consultation by the intervention group midwives was available for 909 women, with general practitioner referral recorded in 268 (29%).

Discussion
Our results show that the redesigned community postnatal care was associated with positive psychological health outcomes in women at 4 months postpartum, although physical health measures did not differ. For the SF36 mental component score there was a 303 point improvement and for the EPDS the odds of a score suggestive of depression relative to controls was 057. The effects were across clusters and not attributable to a few clusters with atypical scores. Some of this improvement in psychological health could probably be explained by the early postpartum detection and management of emotional disorders.21,22 The benefit was not, however, restricted to women at risk of depression but was seen at all levels of mental wellbeing. Although the physical health measure did not differ, the effect could also have been mediated through management of physical health problems. Such disorders are rarely life threatening, but women report an adverse effect on quality of life,4,23,24 and Brown and Lumley25 have shown a relation between physical postpartum symptoms and depression. Although some physical disorders are treatable, many are not amenable to immediate or

016 679 702 160 153 489 522

0689 0009 0008 0206 0216 0027 0022

*Values are odds ratios (95% CI). Values are coefficient (95% CI). Mean reference value=1545. Mean reference value=1106. Mean reference value=877.

Table 5: Multi-level model of women's views about care showing effect of study group

THE LANCET Vol 359 February 2, 2002 www.thelancet.com

383

For personal use. Only reproduce with permission from The Lancet publishing Group.

ARTICLES

Clusters I/C Midwife visits From midwife records From womens diaries GP visits From womens diaries From 4-month questionnaires 17/18 17/19 17/19 17/19

Intervention* 600 (117) 582 (109) 057 (037) 060 (038)

Control* 407 (137) 599 (102) 088 (063) 078 (055)

Difference (95% CI) 192 (104 to 280) 017 (089 to 054) 031 (067 to 004) 018 (051 to 014)

p <00001 0627 0084 0258

I=intervention. C=control. *Values are mean of cluster means (SD).

Table 6: Number of home visits by midwife and general practioner (GP) from various sources

complete resolution, but supportive discussion and reassurance could also have benefited emotional wellbeing. A three-point increase in MCS might not sound substantial, but it is close in magnitude to the differences associated with chronic lung disease, rheumatoid arthritis, and vision impairment (25 to 3 points).17 The direction of effect for MCS was the same as for the EPDS and this translated into a much higher proportion of women with scores suggestive of depression. The effect sizes were similar to those shown in this study for high social support, which is well documented as a predictor of psychological wellbeing.26,27 Some of the assumptions made in calculating the power of this study proved to be underestimates (eg, recruitment and follow-up rate), and others were over estimates (eg, intraclass correlation coefficient). The difference in MCS score was larger than the cautious estimate that had been used in the power calculations, and we were thus able to measure the differences with narrow CIs. Other trials2830 of postnatal interventions aimed at improving womens physical and or psychological health have not shown any benefit. The content of these interventions, an early postnatal check with the general practitioner,28 antenatal and postnatal support groups for high risk women,29 home-based support workers,30 and community-based support groups, information packs, or both31 were all less wide-ranging than the new model of care in this study. Two further trials32,33 of supportive and educational programmes aimed at preventing postnatal depression in women antenatally assessed as high risk had conflicting results. One32 recorded a reduction in EPDS score, but only among the first births and with small numbers (21 vs 24) and non-randomised controls, whereas the other larger, properly randomised trial33 showed no effect. In all but one of the previous trials the interventions were additions to routinely available care, and had low compliance. The intervention in our trial, as a revision of routine UK National Health Service care, was delivered as standard to all women in the group. From trial documentation, most midwives in the intervention group recorded care plans and visits as part of care, regular use of the guidelines, and the completion of symptom checklists, suggesting that the package was implemented with reasonable fidelity. Assessments by the women of the planning of visits, discussion of health symptoms and quality of care relative to expectations, were more positive in the intervention group than in controls. The improvement in mental health we recorded is consistent with the results34 of a trial of midwife-managed maternity care for women of low obstetric risk, in which obstetric interventions were the primary outcomes.35 The main care changes were to antenatal, intrapartum, and hospital postnatal care, with continuity of midwife the only change to the community postnatal care of the intervention group. The midwives who gave the midwife-

managed care in this trial were volunteers, so results might not be generalisable, and there was differential follow-up between intervention and control groups at 7 weeks postpartum, with the possibility of bias. Nevertheless, improvement in the EPDS depression scores of women who had midwife-managed care34 were of a similar magnitude to those in our study. Various issues relating to the validity of results need to be considered in any randomised trial, and those with a cluster design have special potential difficulties, recruitment bias being the most important.36 The random distribution of clusters avoids allocation bias but where, as in this case, staff recruiting patients are aware of the difference between groups, they might form a view that the new model is better. They then might try harder to recruit so that for example, more women with a tendency to be resistant to inclusion could consent in the intervention group, resulting in imbalances. The small imbalances shown between groups (table 2) were such that generally the intervention group did have slightly more women with characteristics compatible with poorer health (thus possibly harder to recruit). However, even if this does reflect recruitment bias, the direction of the bias would decrease rather than increase the likelihood that it could explain the positive mental health effects in the intervention group. A possible differential Hawthorne effect, with intervention midwives more enthusiastic in care and recruitment, was recognised at the design stage and measures taken to keep this to a minimum. Loss to follow-up in the study was just over 20% and non-responders differed in sociodemographic but not obstetric characteristics, a similar pattern to other postpartum studies.1,25 However, such loss is unlikely to have biased outcome comparisons since the rate and pattern of response were close in each group. Women in the intervention group completed the EPDS twice as a screening tool before its use as an outcome measure. With some scales a practice effect can arise with repeated use, although this is not documented for the EPDS. We recorded no evidence of a consistent reduction in scores in the intervention group and the positive effect was in line with the other mental health measure, used for the first time at 4 months. The multifaceted model of care tested in this trial was implemented as a package, therefore it is not possible to know if any of the elements might be more effective than others. Policy makers must be aware that incorporating only part of it might not have the same effect. In deciding whether to adopt the model as standard practice, assessment of resource use is also important and a full cost-effectiveness analysis of all midwife, general practitioner, and secondary care use during the 12 months after birth will be undertaken, with appropriate sensitivity analyses, and reported later. Although the frequency of home visits by general practitioners was significantly lower in the intervention group than in controls, the difference was small. Remuneration of intervention practices did not depend on making home

384

THE LANCET Vol 359 February 2, 2002 www.thelancet.com

For personal use. Only reproduce with permission from The Lancet publishing Group.

ARTICLES

visits or the final check for the trial period, but some general practitioners were reluctant to relinquish these. Midwifery postnatal workload was difficult to measure among control midwives without changing the care they gave and to obtain equally valid measures from both groups was therefore difficult. The data obtained by the midwives suggested 19 fewer visits per woman in controls than in the intervention group, but we had expected under-recording. Although the response rate for womens diaries among controls was lower than in the intervention group, midwife visit rates were close between groups and were consistent with the rate of 66 for England.37 The balance of evidence therefore suggests that the number of visits probably did not differ much, apart from completion of the maternity discharge consultation, which was an additional visit. Visit duration, only recorded by midwives, was on average, 11 min longer for the first, and 5 min for subsequent visits in the intervention group than in controls. At least some of this could be accounted for by the more comprehensive documentation required in the intervention group, especially for the first visit. Midwife familiarity might reduce duration and some items recorded for the trial would be unnecessary if care was standard. Because tailored care can allow midwives to make fewer visits to women with fewer problems, the extended care could well be feasible within present service provision. Compared with standard care our results suggest that womens psychological wellbeing could be significantly improved, and depression reduced at 4 months after birth in women who had redesigned care.
Contributors
The trial was thought of by C MacArthur, H Winter, and D Bick, who, with R Lilford, C Henderson, and H Gee, designed the study and obtained funding. The data collection tools and training models were designed by C MacArthur, D Bick, R Lilford, H Winter, H Knowles, and C Henderson. The guidelines were prepared by D Bick, C MacArthur, H Knowles, and H Winter. Trial implementation was mostly done by D Bick and H Knowles, overseen by C MacArthur, H Winter, and the trial steering committee. Analysis was undertaken by H Winter, R Lancashire, C MacArthur, and D Braunholtz. The results were interpreted by them with D Braunholtz and R Lilford. The article was drafted by C MacArthur and critically revised by all contributors.

7 8 9 10 11 12 13 14

15 16 17

18

19

20

21

22

23 24 25

26

27 28

Conflict of interest statement

None declared. 29

Acknowledgments
The study was funded by the UK National Health Service Research and Development HTA Programme. We thank the midwives, practices, and women who took part in the study, Yvonne Carter for help in design of the study, Anne Gillies for assistance in general practice recruitment and documentation, and Anne Walker for data management and general support. 30

31

32

References
1 2 MacArthur C, Lewis M, Knox EG. Health after childbirth. London: HM Stationery Office, 1991. Glazener CM, Abdalla M, Stroud P, Naji S, Templeton A, Russell IT. Postnatal maternal morbidity: extent, causes, prevention and treatment. Br J Obstet Gynaecol 1995; 102: 28287. Brown S, Lumley J. Maternal health after childbirth: results of an Australian population based survey. Br J Obstet Gynaecol 1998; 105: 15661. Saurel-Cubizolles MJ, Romito P, Lelong N, Ancel PY. Womens health after childbirth: a longitudinal study in France and Italy. BJOG 2000; 107: 120209. Garcia J, Renfrew M, Marchant S. Postnatal home visiting by midwives. Midwifery 1994; 10: 4043. Sharif K, Clarke P, Whittle M. Routine 6-week postnatal vaginal examination: to do or not to do? J Obstet Gynaecol 1993; 4: 25152. 33

34

35 36

5 6

37

Marsh J, Sargent E. Factors affecting the duration of postnatal visits. Midwifery 1991; 7: 17782. Marchant S, Garcia J. What are we doing in the postnatal check? Br J Midwif 1995; 3: 3448. Winterton Report. House of Commons select committee. V1. London: HM Stationery Office, 1992. Cumberlege Report. The expert maternity group. Changing childbirth. London: HM Stationery Office, 1993. Audit Commission. First class delivery: improving maternity services in England and Wales. London: Audit Commission, 1997. Royal College of Midwives. Life after birth: reflections on postnatal care. London: Royal College of Medicine, 2000. Altman DG, Practical statistics for medical research. London: Chapman and Hall, 1991. Bick DE, MacArthur C, Knowles H, Winter HR. Postnatal Care: evidence and guidelines for management. London: Harcourt Brace, 2001. Oakley A. Experiments in knowing. London: Polity Press, 2000: 18284. Townsend P, Phillimore P, Beattie A, et al. Health and deprivation. Inequality and the North. London: Routledge, 1988. Ware JE, Kosinski M, Keller SD. Physical and mental health summary scales: a users manual. Boston: The Health Institute, 1994. Cox JL, Holden JM, Sagovsky R. Detection of postnatal depression: development of the 10 item postnatal depression scale. Br J Psychiatry 1987; 150: 78286. Premarantne UN, Sterne JAC, Webb J, Burney PGJ. A general practice based community intervention on the management of asthma. Thorax; 52: A18. Rasbash J, Browne W, Goldskin H, et al. A users guide to MLWin, Multilevel models Project. London: Institite of Education, University of London, 2000. Ray KL, Hodnett ED. Caregiver support for postpartum depression (Cochrane Review). In: The Cochrane Library, Issue 4. Oxford: Update Software. 2000. Clement S, Elliot S. Psychological health before during and after childbirth. Ch 18. In: Marsh G, Renfrew M, eds. Community-based maternity care. Oxford: Oxford University Press, 1999. Bick DE, MacArthur C. The extent, severity and effect on health problems after childbirth. Br J Midwif 1995; 3: 2731. Trevalyan J. Please tell mother. Nursing Times 1994; 90: 3839. Brown S, Lumley J. Physical health problems after childbirth and maternal depression at six to seven months postpartum. Br J Obstet Gynaecol 2000; 107: 1194201. Oakley A, Hickey D, Rigby AS. Love or money? Social support, class inequality and the health of women and children. Eur J Public Health 1994; 4: 26573. Oakley A. Social Support and Motherhood. Oxford: Blackwell, 1992. Gunn J, Lumley J, Chondras P, Young D. Does an early postnatal check-up improve maternal health: results from a randomised controlled trial in Australian general practice. Br J Obstet Gynaecol 1998; 105: 99197. Stamp GE, WIlliams AS, Crowther CA. Evaluation of antenatal and postnatal support to overcome postnatal depression: A randomised, controlled trial. Birth 1995; 22: 13843. Morrell CJ, Spiby H, Stewart P, Walters S, Morgan A. Costs and effectiveness of community postnatal support workers: a randomised controlled trial. BMJ 2000; 321: 59398. Reid M, Lang G, Murray G, Prigg S, Glazener C. A two centre pragmatic randomised controlled trial of two interventions. Edinburgh: Scottish Office Chief Scientist Office, 1999. Elliott SA, Leverton SJ, Sanjack M, et al. Promoting mental health after childbirth: a controlled trial of primary prevention of postnatal depression. Br J Clin Psychol 2000; 39: 223241. Brugha TS, Wheatley S, Taub NA, et al. Pragmatic randomized trial of antenatal intervention to prevent post-natal depression by reducing psychosocial risk factors. Psychol Med 2000; 30: 127381. Shields N, Reid M, Cheyne H, et al. Impact of midwife-managed care in the postnatal period: an exploration of psychosocial outcomes. J Reprod Inf Psychol 1997; 15: 91108. Turnbull D, Holmes A, Shields N, et al. Randomised controlled trial of efficacy of midwife-managed care. Lancet 1996; 348: 21318. Okoumunne OC, Gulliford MC, Chinn S, Sterno JAC, Burney PGJ. Methods for evaluating area-wide and organisation-based interventions in health and health care: a systematic review. Health Technol Assess 1999; 3 (5): 197. Department of Health. Community Maternity Services Summary Information for 199798. England: Government Statistical Service.

THE LANCET Vol 359 February 2, 2002 www.thelancet.com

385

For personal use. Only reproduce with permission from The Lancet publishing Group.