Stability of Maternal Preferences for Pediatric Health States in the Perinatal Period and 1 Year Later FREE

ALTHOUGH SURVIVAL of extremely premature infants has improved dramatically in recent years, they remain at high risk for subsequent neurological impairments.^1- 2 Concerns about the possibility of a further increase in the prevalence of impairments have increased with the lowering of the threshold of viability.^3- 4 Many parents, health care professionals, and medical bodies are now questioning whether intensive care is justified for all infants, regardless of birth weight.^5- 9 As a result, physicians are increasingly inviting parents to share in the responsibility of decision making in the perinatal period.^4,7,10 These decisions are likely to have profound immediate and long-term consequences for the family, the health care system, and society.

However, parents typically have very little understanding of the extent of uncertainty and the types of expected future outcomes for extremely preterm infants. Even when parents are involved, decisions to initiate or forgo intensive care are often made arbitrarily in the delivery room, without presenting decision alternatives, and with little insight regarding parental preferences for future outcomes. Preference in this context is used as an expression of "serious deliberation" in which values are assigned to different outcomes after a "sincere attempt to consider alternative choices of action."¹¹ Several studies indicate that health care professionals and parents often differ in their valuation of disabling health states and in their attitudes toward saving infants of borderline viability.^12- 15 To date, very little information is available regarding parental perspectives in terms of the quality of life of their children. We have shown that although parents of adolescent premature children (and the children themselves) provided relatively high valuation for their children's health states, despite recognition of significant functional deficits,^16- 17 a substantial proportion provided lower scores than the control group. However, a number of years have elapsed since they experienced the crisis of premature birth. The question therefore remains whether the preferences collected at adolescence represent preference scores of parents had they been measured in the perinatal period. The assumption that preference scores for future health states are stable is fundamental to using the information on parental preferences when making clinical decisions in the delivery room. This assumption has not been tested, particularly in the context of childbirth and infancy.

The objectives of this study were to determine whether preference-based health-related quality-of-life scores obtained serially from women with high-risk pregnancies and mothers of newborn very low-birth-weight (VLBW) infants were stable from the perinatal period through to 1 year of corrected age (CA). Preferences of women in the antenatal period are largely unknown, and it is not clear whether they differ from those obtained after the birth of an infant. In addition, the VLBW cohort was recruited because we anticipated that only a small proportion of women with high-risk pregnancies were likely to be delivered of premature infants or infants at high risk for subsequent neurodevelopmental problems. We hypothesized that preferences would be stable over time and that selected maternal sociodemographic and emotional factors, infant variables at birth, and subsequent health and neurodevelopmental problems would not affect maternal preference scores.

A cohort of consecutive women with very high-risk pregnancies attending a tertiary care regional perinatal center clinic was identified by an obstetrician using predefined criteria (Ontario Ministry of Health Prenatal Form). We included women with incompetent cervix; women with previous history of stillbirth, neonatal death, or premature birth; and women carrying fetuses with congenital chromosomal malformations. All non–English-speaking mothers were excluded. Eligible women were approached for recruitment by a trained research assistant (B.L.S.) at antenatal visits at 24 ± 2 weeks of gestation.

This cohort of mothers of infants who weighed less than 1500 g at birth was recruited from consecutive admissions to the tertiary care neonatal intensive care unit at McMaster Children's Hospital, Hamilton, Ontario. Mothers of 26 infants who were facing imminent death were not approached on compassionate grounds; all of these infants died. Mothers of eligible infants were approached for consent at an approximate postnatal age of 5 to 7 days. Although it would have been desirable to approach the mothers as close to the birth of the infant as possible, our pilot experience indicated that the women would not have been receptive to this complex series of interviews. None of the respondents in either group was paid for their participation.

The baseline characteristics of eligible candidates who refused to participate were recorded to determine whether the study sample was representative of the population.

The study was approved by the Ethics Committee of Hamilton Health Sciences, McMaster University, Hamilton. Written informed consent was obtained from all respondents.

Information on maternal demographics, socioeconomic status (SES),¹⁸ pregnancy history, pregnancy risk category, history of infertility, planned pregnancy, and previous maternal exposure to children with disabilities was obtained at recruitment. Global assessment of maternal health (excellent, very good, good, fair, or poor) and that of her spouse and other children, and current maternal health status were obtained at each visit using the Health Utilities Index (HUI2) classification system.¹⁹

On the basis of the premise that maternal anxiety, depression, or lack of social support may have a bearing on the responses provided for the hypothetical health states, all women completed the Spielberger State-Trait Anxiety Inventory for Adults,²⁰ which has 2 components. The State Inventory measures how one feels at present, and the Trait Inventory measures how one feels generally. They also completed the Center for Epidemiologic Studies Depression Scale²¹ and the Sarason Social Support Questionnaire,²² which measures the number of people one depends on and how satisfied one is with the support. These self-completed questionnaires are well validated and widely used in studies in the perinatal period.^23- 24

Preference Measurement Techniques. We used the same preference measurement techniques as in previous studies from our group²⁵ to quantify the health-related quality-of-life scores of 5 pediatric hypothetical health states. The visual analog scale (Feeling Thermometer) was administered first to introduce the respondents to the concept of preferences. The Standard Gamble (Chance Board) involves a lottery approach to measurement and provides a single cardinal utility score ranging from 0 to 1.00, where 0 indicates death and 1.00 is perfect health. This method is more accepted for measurement of preferences, particularly where uncertainty of outcomes is an important characteristic of the decision,²⁶ and therefore only Standard Gamble data will be presented herein.

Hypothetical Health States. Five hypothetical health state descriptions based on health states reported for extremely low-birth-weight children in the 8-year study^27- 28 were selected for evaluation (Table 1). These descriptions were based on the HUI2 system, which has 6 attributes, each with 3 to 5 functional levels.¹⁹ These attributes were sensory perception, mobility, emotion, cognition, self-care, and pain. Each health state was given a unisex name, ie, Jamie, Chris, Alex, Sandy, and Pat.¹³ The health scenarios ranged from a mild single condition (Jamie and Alex) to moderate (Chris) and very severely disabling health states (Sandy and Pat). The health state descriptions were color coded for severity and presented to the respondents in random order on 4 × 6-in cards. Respondents were asked to imagine themselves living in each of the above states of health for the next 60 years. As in previous studies, we did not ask parents to value the life of their children in such health states because of the emotive responses this task might elicit. Further details of the methods and descriptions of the hypothetical health states are available in previous publications from our group.^13,16

Severity of illness was measured on the first day of life using scores on the Clinical Risk Index for Babies measure.²⁹ Data on birth characteristics and all neonatal problems were recorded from the medical chart after infant discharge. The mother's perception of her infant's severity of illness was obtained by asking "How ill is your child now?" and using a 7-point Likert scale (ranging from "not at all ill" to "extremely ill") at the postnatal interview.

Both cohorts of infants were seen at the follow-up clinic at a CA of 12 months. The assessments were performed by a pediatrician (S.S.) and included anthropometric measurements, global assessment of the infant's health, a physical and neurological examination, and clinical judgment of the infant's developmental age. At the same visit, a research coordinator administered the Vineland Adaptive Behavior Scales³⁰ to the mother. The Vineland Adaptive Behavior Scales, a maternal report measure, was chosen because maternal preferences are more likely to be influenced by the mother's perception of her infant. Additional information on the infant's general health, illnesses, surgery, hospitalizations, medications, vision and hearing problems, special needs, and use of special resources, and maternal concerns regarding the infant's health and development was obtained in the same visit.

The interviews were conducted in a private room at the Health Sciences Centre of McMaster University by highly trained lay professional interviewers who were masked to the medical history and the hypotheses being tested. All interviews were tape-recorded for quality control. We reviewed a random sample of the tapes to ensure strict adherence to the study protocol. The preference interview took approximately 1 hour to complete. All mothers were interviewed on the following 2 occasions: 7 to 10 days after birth and at their infant's 12-month CA visit. The pregnant women (antenatal cohort) were also interviewed at 24 ± 2 weeks' gestation.

Estimates of parameters for sample size calculations were obtained from results of reported utility measures for 3 hypothetical health states collected by Feeny et al.^31- 32 The mean of the SD of differences was 0.247, and the mean correlation was 0.59. The mean minimum clinically important difference in utility scores for the same hypothetical health state at different assessment points was preset at 0.075 or greater, and power was set at 80% (α = .05). Using these assumptions, we calculated the minimum required number of individual subjects with paired observations to be 63.

Before statistical analysis, health state preference scores were transformed to scales that would take into account states rated as worse than death. The Scale for Transformed Utility Scores (Standard Gamble) was defined from 1.0 to −1.0, using the nonlinear transformation y = x/(1 − x), described by Patrick et al.³³ Statistical analyses were conducted using SPSS software (Version 10.05; SPSS Inc, Chicago, Ill). We analyzed changes in preference scores among antenatal, postnatal, and 12-month follow-up visits using multivariate repeated-measures analysis of variance. We used the following model for the main analysis:

2(Time − Within Subject) × 2(Method − Within Subject) × 5(Health States − Within Subject).

The multivariate analysis of covariance was also used to determine whether infant factors (birth weight [≤1000 or >1000 g], duration of mechanical ventilation, health problems at 1 year, and visits to the emergency department) or maternal factors (educational level, history of pregnancy problems, high-risk pregnancy, anxiety/depression, and social support) influenced preferences for health states. For this multivariate analysis of covariance we used the following model:

2(Time – Within Subject) × 2(Method –Within Subject) × 5(Health States – Within Subject)

with the maternal and infant variables as covariates.

All maternal and infant variables were summarized using descriptive statistics. We compared repeated measures of continuous variables using the t test (paired). We measured interrater reliability between maternal and physician assessments of the child's health using the κ statistic. We measured intrarater reliability of maternal responses to health assessments (ie, global health, satisfaction with social support) repeated at each visit using the Spearman rank correlation. We used multivariate linear regression analysis to test whether individual health state scores (dependent variables) in the VLBW cohort were affected by maternal and neonatal variables (independent variables).

We compared participants and those who refused to participate using the t test for independent groups to compare mean differences in means, Mann-Whitney test to compare differences in medians, and χ² analysis to compare differences in proportions. Multivariate logistic regression was conducted to identify factors that predicted a mother's consent or refusal to participate in the study.

During the recruitment period from March 1, 1998, through February 28, 1999, 80 (57%) of 140 eligible women consented to participate in the study. Most of the women who refused cited busy schedules and multiple medical appointments as the reason. The women who refused had a significantly higher pregnancy risk status (P<.001) and a lower SES (P = .002). There were no differences in their educational status or other demographic variables (maternal education and SES were available for only half of those who refused in both cohorts). By 12 months, 73 (91%) of 80 women completed all 3 interviews.

Seventy-five (67%) of 112 eligible mothers of VLBW infants consented to participate in the study. Eleven women (30%) who refused were too busy with other responsibilities to participate; 11 (30%) were too overwhelmed and stressed; and the rest had other pressing reasons. Mothers who refused were younger (P = .03), less educated (P = .007), and in a lower SES class (P = .002) than those who consented. We found no significant differences in mean gestational age between infants of mothers who refused to participate compared with those who participated (nonparticipants, 28.5 weeks [SD, 2.8 weeks]; participants, 29.2 weeks [SD 2.5 weeks]; P = .16). We also found no differences in mean scores from the Clinical Risk Index for Babies between nonparticipants and participants (nonparticipants, 2.9 [SD, 3.1]; participants, 2.0 [SD, 2.0]; P = .09). By 12 months, 72 (96%) of 75 participants in the cohort completed both interviews.

Maternal demographics for both cohorts showed predominantly white, Canadian-born women of middle-to-upper SES in a 2-parent family relationship (Table 2). Other than a slightly higher SES, the rest of the demographic characteristics are representative of the regional population.

Antenatal Cohort. As expected, a high proportion of infants (n = 48; 56%) were born at term. Among the infants, 9 (10%) had a birth weight of no greater than 1500 g; 1 (1%) weighed less than 1000 g; 5 (6%) had congenital abnormalities; and 6 (7%) were multiple births, resulting in 86 infants (Table 2). The mean Clinical Risk Index for Babies score was 0.2 with few neonatal complications; 17 infants (20%) underwent mechanical ventilation, and the mean hospitalization was 13 days (range, 2-41 days).

VLBW Cohort. The 21% multiple birth rate resulted in 93 infants born to 75 mothers. This cohort was moderately ill, with a mean Clinical Risk Index for Babies score of 2.0, and 80 infants (86%) who underwent mechanical ventilation. The mean duration of mechanical ventilation, including continuous positive airway pressure, was 21 days (SD, 29 days), and the mean length of hospitalization was 59 days (range, 24-131 days).

Antenatal Cohort. Spielberger Trait scores were measured only once at 24 weeks of gestation for the antenatal cohort and at the first interview only for the VLBW cohort; the scores were in the middle range for both cohorts. Maternal State anxiety and depression scores decreased significantly from the antenatal period to the 12-month visit (Table 3). Spielberger State anxiety scores did not decrease significantly from the antenatal to postnatal visits or the postnatal to 12-month visits; however, we found a mean decrease of 5 points between the antenatal and 12-month visits (95% confidence interval [CI], –7.2 to –2.8; P<.001). Mean depression scores did not differ significantly between the antenatal and postnatal visits. However, we found significant decreases in scores between the antenatal and 12-month scores (mean difference, −2.8; 95% CI, –5.1 to –0.7; P = .03), and between the postnatal and 12-month scores (mean difference, −5.3; 95% CI, –8.2 to –2.6; P<.001). We found no change in mean scores for satisfaction with social support across the 3 assessment times.

VLBW Cohort. Maternal State anxiety and depression scores decreased significantly from the initial postnatal period to the 12-month follow-up visit (Table 3). Maternal State anxiety scores decreased by an average 8.6 points (95% CI, −11.6 to −5.6; P<.001), and maternal depression scores declined by an average of 9.5 points (95% CI, −12.1 to −7.1; P<.001). Again, measures of social support did not change over time.

Eighty-one (94%) of 86 infants were seen at the 12-month visit (Table 4). Most mothers and clinicians (>85%) rated the infants' health as excellent or very good, and there was moderate agreement between these assessors (κ = 0.48); 75 mothers (93%) had no concerns about their infant's development. Almost half the infants had experienced some illness in the first year of life (n = 39) and had visited the emergency department (n = 32); only 7 (9%) had experienced overnight hospitalization. Approximately one third of infants had required visits to medical specialists, but used few other special resources. All infants had normal neurological findings except for 1 infant who required a hearing aid. Clinical assessment of mental and motor developmental age concurred with age at the visit. The Vineland Adaptive Behavior Scales scores and composite scores were all within the reference range.

Ninety (97%) of 93 VLBW infants were seen at the 12-month CA visit (Table 4). Global health assessment was excellent to very good for most infants as judged by the mothers (n = 75; 83%) and clinicians (n = 66; 73%), and agreement was good between these assessors (κ = 0.49). Nine mothers had concerns about their infants' delayed motor development, and 1 mother for motor and mental delays. More than half of the VLBW infants (n = 50; 56%) had experienced some illness in the first year of life; 18 infants (20%) had been hospitalized overnight; and emergency department visits were common. Use of special resources was high (73 infants [81%]), predominantly with visits to medical specialists, followed by use of infant stimulation programs and physiotherapy or occupational therapy services. Most infants had normal physical and neurological assessments. Only 5 infants had neurological problems. Three infants wore glasses and 1 had a hearing aid. The clinicians' mean ratings of the infants' developmental age of 11.7 months (SD, 1.2 months) for mental abilities and 11.4 months (SD, 1.6 months) for motor skills were significantly lower (P<.001) than the actual mean CA of 12.2 months at the time of assessment. The Vineland Adaptive Behavior Scales showed scores within the reference range for each dimension.

Mothers' preferences for individual health states were compared at 3 time periods (Figure 1). Standard Gamble ratings were affected by differences in health states (F_4,288 = 87.0; P<.001), but did not change over time (F_2,144 = 1.3; P = .29). Intraclass correlations (r = 0.99) indicated high agreement among participants in the strength of their preferences for individual health scenarios. Thus, in general, health scenarios that described milder disabilities were consistently given the highest ratings (Jamie and Alex). The scenario describing moderate disability (Chris) was consistently assigned intermediate preferences scores, and scenarios with severe disabilities (Sandy and Pat) were consistently given the lowest ratings. The variability in scores was greater for the most severely disabling health states with wide SDs for both cohorts (Table 5). The proportion of women who rated 1 or more health states to be equal to or worse than death was not different across the 3 time periods (36/80 [45%], 29/76 [38%], and 30/73 [41%] for the antenatal, postnatal, and 12-month CA visit, respectively).

Mothers' preferences for individual health states at the infants' 12-month CA visit were compared with their initial preferences measured at the postnatal visit (Figure 1). As with the antenatal cohort, Standard Gamble ratings were affected by differences in the severity of health states (F_4,284 = 64.2; P<.001), but did not change over time (F_1,71 = 0.7; P = .42) (Table 5). Intraclass correlations (r = 0.95) indicated high agreement among participants in the strength of their preferences for individual health scenarios. The proportion of women who rated 1 or more health states to be equal to or worse than death was not different between the 2 time periods (35/72 [49%] vs 32/72 [44%]).

Repeated-measures analysis showed no differences between the 2 cohorts in preference scores for the 5 hypothetical health states obtained at the postnatal and 12-month follow-up visits (F_1,143 = 0.54; P = .46).

Multivariate repeated-measures analysis of variance and regression analysis were used to assess the effects of selected neonatal and maternal variables on the stability of preference scores obtained longitudinally. To deal with the problem of statistical analysis arising from the inclusion of multiple births, a computer-generated table of random numbers was used to select one of the twins or one of the triplets for inclusion.

The stability of preference scores in both cohorts was not affected by maternal education level (antenatal cohort, F_1,71 = 0.6 [P = .4]; VLBW cohort, F_1,70 = 3.0 [P = .09]) or a poor outcome in previous pregnancy (antenatal cohort, F_1,71 = 2.0 [P = .2]; VLBW cohort, F_1,70 = 2.5 [P = .1]). Regression analysis on the impact of depression, anxiety, social support, maternal perception of her own global health, and maternal perception of the infant's severity of illness did not show any effect on maternal preference scores for hypothetical health states for either cohort (P>.05 for all).

Maternal Standard Gamble preference scores for either cohort were not affected by birth weight (antenatal cohort, not applicable because birth weight in only 1 infant was <1000 g; VLBW cohort, F_1,70 = 0.5 [P = .5]), duration of mechanical ventilation (antenatal cohort, F_1,71 = 0 [P = .97]; VLBW cohort, F_2,69 = 1.5 [P = .2]), infant health problems (antenatal cohort, F_1,71 = 0.7 [P = .4]; VLBW cohort, F_1,70 = .01 [P = .9]), or emergency department visits (antenatal cohort, F_1,71 = 1.2 [P = .3]; VLBW cohort, F_1,69 = 0.14 [P = .7]).

The clinical acuity and emotional intensity around the period of birth and the first few days of life for infants of borderline viability make decision making at this point particularly difficult, complex, and challenging for parents and health care professionals. The recent increasing involvement of consumers and parents in this joint endeavor requires us to have a better understanding of parental preferences.^4,7,10 However, limited information exists about how parents view the quality of life for different health outcomes in childhood, and we have no knowledge of whether parental preferences at critical moments in the perinatal period are stable over time, or whether other factors such as life experiences influence recalibration of them.¹⁷ In this study, therefore, we sought to answer these questions using real-life players (mothers who experienced high-risk pregnancies or who were delivered of high-risk infants), mimicking real-life clinical practice (discussing probabilities of different health outcomes at critical phases of family experiences), and using realistic scenarios (hypothetical health states derived from actual clinical states experienced by premature infants).

We have shown that maternal preferences for health states measured over time using the Standard Gamble technique were stable between the antenatal period, immediate postnatal period, and up to 1 year later. Preference scores provided by the 2 cohorts of women for each hypothetical health state tended to mirror one another. Preference scores were affected only by the severity of the individual health states. The ratings appear realistic for the type of the health state (ie, severely disabling health states were consistently given the lowest scores). The 2 most disabling health states (Pat and Sandy) elicited the most variability in scores, and nearly one third of the mothers consistently rated 1 or more of these health states as worse than death. The mean change in scores between assessments for the same health states did not exceed the preset minimally important difference of 0.075 in either cohort (except for 1 value for Pat in the antenatal cohort, which may be a chance finding). Overall, the scores for the individual hypothetical health states are remarkably similar to those provided by other respondents (adolescents born prematurely or at term, their parents, and health care professionals) in previous studies.^13,16- 17

We know very little about the factors that shape and influence preferences.^{11,13,16- 17} In this study, we explored the effects of a wide array of maternal and infant variables on stability of preference scores for hypothetical health states. Variables were chosen on the basis of literature review and our collective clinical judgment. Despite higher levels of anxiety and depression around the birth of the infant and the subsequent decline of these levels a year later, none of the maternal or infant variables appear to have had any impact on the change in preference scores. Because of the small number of infants with neurological impairments, we were unable to answer the question of whether personal experience with disabilities in this pregnancy had any effect on preference scores. Nevertheless, we believe that knowledge about the stability of maternal preferences over time at different critical phases through pregnancy, childbirth, and the first year are useful.

A complex study such as this, which requires a commitment to lengthy serial interviews at extremely stressful periods in the life of women, is very difficult to conduct. This is reflected in the less-than-optimal consent rate of subjects during recruitment (despite our vast experience in this field), where at least one third of women declined to participate. However, once they agreed to participate, compliance and retention were excellent. Although the reasons for nonparticipation were generally lack of time or other priorities, the participants were somewhat better educated and of a higher social class than the nonparticipants. We recognize that this bias in ascertainment may limit the generalizability of our findings. Despite our best efforts, we found that it was virtually impossible to assess all subgroups of interest. We are somewhat reassured by the report that little of the variability in preference scores is explained by sociodemographic characteristics of respondents.³⁴

Another limitation of our study is that preferences were elicited only from mothers. Our reliance on the mothers is in no way meant to undermine the importance of fathers in decision making. However, logistic reasons (we anticipated that compliance by fathers would be even more challenging) and funding constraints precluded us from including both sets of parents. Although the literature is sparse in this area, analysis of data from 3 surveys showed that the sex of parents is not an important factor in explaining interrater variability in preference scores.³⁴ Another study showed important differences in mean scores among families, but not between parents within a family.³⁵ Future studies in this field should incorporate preferences from both parents, parents whose infants die, parents of different ethnic groups and social classes, and parents of children with severe disabilities.

A literature review suggests that few studies exist in which preferences have been obtained from parents in the perinatal period.^32,36- 38 Stability of preferences, however, has been studied extensively with regard to life-sustaining treatments among the elderly. These studies showed that most patients who chose to forgo life-sustaining treatments did not change their choices with time and generally supported the use of advance directives,^39- 43 whereas others recommended periodic reassessments of patients for life-sustaining care.^44- 45 Several other investigators have shown that even in the face of changes in health status, the evaluations in patients with cancer and human immunodefiency virus infection remained stable throughout the course of treatment.^46- 50 The findings of stability of preferences of patients over time are reassuring if such models are to be incorporated into decision analysis. However, Llewellyn-Thomas et al⁴⁹ caution against generalization of their findings to other clinical contexts.

To our knowledge, this is the first time that maternal preferences regarding disabling health states have been obtained in the perinatal period. Despite some limitations, our study adds substantially to the understanding of preferences at critical moments in the life of parents. Although parental preferences may be in a state of evolution during the course of pregnancy and after childbirth, we have shown that even under conditions of stress and anxiety, preferences are remarkably stable during this period and 1 year later. The absence of significant variation in maternal values by sociodemographic, neonatal, and intrapsychic factors also speaks to the likelihood that these values and preferences are relatively robust and are not subject to external forces that might in the past have been assumed to be important determinants of these values. In an era of family-centered care, it has become important, within the bounds of ethical practice, to solicit the perspectives and listen to the wishes of parents and families.⁵¹ Collaborative decision making is increasingly advocated as an ideal model in life-threatening situations and where a high probability of an ambiguous outcome exists, as in the case of infants of borderline viability.^15- 16,52 Service providers, therefore, have an increasing obligation to seek, understand, and use parental values and wishes in shared decision making in the perinatal period.

Corresponding author and reprints: Saroj Saigal, MD, FRCPC, Department of Pediatrics, McMaster University, 1200 Main St W, Room 4G 40, Hamilton, Ontario, Canada L8S 4J9 (e-mail: saigal@mcmaster.ca).

Accepted for publication September 26, 2002.

This study was funded by grant MT 14427 from Medical Research Council of Canada, Ottawa, Ontario, and grant XG 98-051 from the Hospital for Sick Children's Foundation, Toronto, Ontario.

We are extremely grateful to the women who participated in this study at very stressful periods in their lives. We would like to acknowledge the contributions of David Feeny, PhD, University of Alberta, Edmonton, and William Furlong, MSc, McMaster University, Hamilton, in the design of this study and for their helpful comments; Barbara Brennan, MD, and the staff of the Antenatal Clinic, and the staff of the neonatal intensive care unit of McMaster University Medical Centre for their assistance in recruitment of the subjects; Lorraine Hoult for her assistance in conducting the study; Liz Merz for tracking the subjects; Victoria Donnelly, MPhil, for data entry and data verification tasks; and Diane Turcotte for typing the manuscript. We appreciate the support of the Department of Pediatrics and McMaster Children's Hospital, Hamilton.

This is the first time, to our knowledge, that maternal preferences regarding disabling health states have been obtained in the antenatal and perinatal periods. We have shown that even under stressful conditions, preferences were remarkably stable during the first year after birth. Service providers, therefore, have an increasing obligation to elicit and consider parental values in decision making.