Patterns of visit attendance in the nurse-family partnership program

Patterns of Visit Attendance in the Nurse–FamilyPartnership ProgramMargaret L. Holland, PhD, MPH, Yinglin Xia, PhD, MS, Harriet J. Kitzman, PhD, Ann M. Dozier, PhD, and David L. Olds, PhD

Low visit attendance is a common problemwith preventive interventions1 and occursacross home visitation programs.2---5 Missedvisits are concerning because they are missedopportunities for providers to educate andsupport families. The positive outcomesintended for these programs are diminishedwith shorter duration of involvement4 and lesscontact time between the family and the visi-tor.6 The degree to which low visit attendanceis a problem, however, is not easily quantifiedbecause there may not be a simple dose---response relationship between visit attendanceand outcomes.

Missed visits are associated with participantbackground characteristics and outcomes in-consistently across studies. Some of these in-consistencies may be attributable to differencesin goals, target populations, and service pro-viders. Other apparent inconsistencies may re-sult from assumptions of linear or dichoto-mized relationships,5,7---10 in which there issome evidence of nonlinear associations. Forinstance, in the Elmira, New York, and Mem-phis, Tennessee, trials of the Nurse---FamilyPartnership (NFP), completing more visits wasassociated with lower maternal psychologicalresources in a linear relationship with an indexbased on mother’s intelligence, mental health,mastery (the extent to which the mother be-lieves she can control her own life outcomes),and self-efficacy11 as well as a quadratic func-tion with mothers with high and those with lowpsychological resources both receiving morevisits than mothers with average psychologicalresources.12

Lower visit attendance has been associatedwith lower income8,10 and education3,8 andworse mother and child health.10 However,stress, substance abuse, and mental healthproblems are associated with higher visit at-tendance.7 In one study, more social supportwas associated with lower visit attendance,7 butin another, being married and living witha partner was associated with higher visit

attendance, whereas living alone was associ-ated with lower visit attendance.3 The rela-tionship between psychosocial characteristicsand visit attendance is complex.

Completed visits require cooperation be-tween the mother and the nurse; the mothermust be available at the scheduled time and thenurse must reach out to build trust and mustreschedule missed appointments. Differencesin visit patterns may reflect differences infamilies’ needs, mothers’ abilities to participate,and visitors’ interpretations of those needs.Visitors’ responses to these factors may differdepending on service provider characteristicsor the personality match between the motherand nurse.13 In one program, sites with visitorswho delivered the program with more flexibil-ity had higher retention than did other sites,3

and increasing visitor flexibility increasedcompleted home visits.2

Visit attendance can be measured in manyways, including time to attrition, number ofvisits completed, and total contact time.

Examining visit patterns over time instead ofthese aggregate measures may uncover non-linear relationships and provide insight into thefamilies’ experiences. Visit patterns can indicateif there are common times when families dropout or frequently miss visits or if there are othercommon trajectories. Previous work identifyingattendance patterns focused on programs witha fixed number of visits,8,14 but home visitingprograms often have a variable number,depending on the timing of enrollment.

Visit attendance patterns at specific timesduring an intervention with a variable numberof sessions have not been studied, to ourknowledge, nor have the relationships betweensuch patterns and outcomes. Identification ofdistinct visit attendance patterns may helppredict which families would benefit from re-tention efforts and may improve our under-standing of complex relationships betweendose and outcomes.

NFP is an evidence-based home visitationprogram implemented nationally15 in which

Objectives. We examined visit attendance patterns in the Memphis trial of the

Nurse–Family Partnership and associations between these patterns and family

characteristics, outcomes, and treatment–control differences in outcomes.

Methods. We employed repeated measures latent class analysis to identify

attendance patterns among the 228 mothers assigned to receive home nurse

visits during pregnancy and until the child was aged 2 years, associated

background characteristics, outcomes, and treatment–control differences by

visit class. Home visits were conducted from June 1990 to March 1994. We

collected outcome data from May 1992 to April 1994 and July 2003 to December

2006.

Results.We identified 3 visit attendance patterns. High attenders (48%) had the

most visits and good outcomes. Low attenders (33%) had the most education

and the best outcomes. Increasing attenders (18%) had the fewest completed

visits during pregnancy, the poorest intake characteristics, and the poorest

outcomes. Treatment–control group differences varied by class, with high and

low attenders having better outcomes on some measures than did their control

group counterparts.

Conclusions. Three patterns were associated with distinct groups of mothers

with different long-term outcomes. Further examination and use of patterns to

classify mothers and prioritize resources may improve efficiency in the Nurse–

Family Partnership. (Am J Public Health. Published online ahead of print August

14, 2014: e1–e8. doi:10.2105/AJPH.2014.302115)

RESEARCH AND PRACTICE

Published online ahead of print August 14, 2014 | American Journal of Public Health Holland et al. | Peer Reviewed | Research and Practice | e1

fewer than half of recommended visits werecompleted in the original trials and in thecurrent national replication of the program.2,3

NFP starts during pregnancy and recommendsvisits until the child’s second birthday.16 Visitfrequency varies over time to support rela-tionship development between the nurse andfamily and to accommodate shifts in maternaland child health and development over this2.5-year period.17 Nurses promote improve-ments in the mothers’ health and parentingbehaviors, economic self-sufficiency, and sup-portive relationships and link mothers withhealth and community services.18

Three randomized controlled trials of theNFP model found consistent improvements ina range of outcomes, including children’s homeenvironments, children’s language develop-ment, childhood injuries, and the timing ofsubsequent pregnancies for mothers.11,19---21

The intervention affected some outcomes forthe whole sample and others (including child’sacademic achievement) only for children ofmothers with low psychological resources(40%---50% of the samples).

The effects of NFP depend on engagement ofthe families with the program in accordancewith the visit patterns achieved in the originalNFP trials. Because of different levels of familyneed and engagement, we examined 3 ques-tions: (1) Were there discernible variations incompleted visit patterns over the course of theprogram? (2) Were those variations associatedwith risk and outcomes? (3) Were thereintervention---control group differences in out-comes for the subgroups defined by visitpatterns?

METHODS

We used data from the Memphis trial of theNFP, which enrolled women in 1990 to 1991.Primiparous women at less than 29 weeks ofgestation were recruited at an obstetric clinicfor the poor. Participants met 2 of 3 criteria:unemployed, unmarried, less than 12 years ofeducation. The resulting sample was primarilyBlack (92%), young (64% younger than 19years), poor (85% at or below federal povertyguidelines), and unmarried (98%).11

The control (n = 515) and intervention (n =228) groups both received transportation toprenatal care, developmental screening, and

referral services for children.18 The interven-tion group was offered nurse home visits fromenrollment until the child was aged 2 years.Home visits were conducted from June 1990through March 1994. Nurses aimed to com-plete all recommended visits but were encour-aged to make greater attempts at engagingmothers with the greatest need.17 Interviewersmasked to treatment assignment collected out-come data periodically after the child’s birthuntil the child was aged12 years. We examinedselected maternal and child outcomes from the24-month (May 1992 to April 1994) and12-year (July 2003 to December 2006) as-sessments. The participants who were inter-viewed at 12 years had characteristics equiva-lent to those of the original participants.21

We excluded participants who experiencedmiscarriages (n = 8), stillbirths (n = 2), andneonatal losses (n = 1) from this study, becausevisits were discontinued. Visit data were miss-ing on 5 mothers. The resulting sample size was212 for visit pattern analyses. Twelve-yearoutcome data were missing on 33 children,resulting in a sample size of 179 for thoseanalyses. The control group sample size was469 for 24-month outcomes and 397 for12-year outcomes.

Measures

We considered all in-person visits completedvisits. The program environment was charac-terized by whether there was racial concor-dance with the initial nurse and whether thenurse assigned to the family changed duringprogram participation, primarily because ofnurse turnover.

We included maternal age as a 3-level cat-egorical variable (< 17 years, 17---18 years,and > 18 years) to allow a nonlinear relation-ship, following the original trial stratification.11

Mothers younger than 17 years are at greaterrisk for having children of low birth weight andbenefited more from the program in the pre-vious, Elmira trial. Those aged 18 years andyounger are at greater risk for a variety of childhealth and developmental problems, includinglimited economic self-sufficiency. Mother’semployment reflected her age, family andeconomic situation, and availability to attendvisits.

Other demographic measures were race,marital status, living alone, attending school,

the grade of school completed, contact betweenthe mother and father of the baby (“fatherinvolved”), receipt of government assistance,household discretionary income (differencebetween self-reported income and subsistencestandards on the basis of household size), andneighborhood poverty (percentage of house-holds at or below the federal poverty level inthe census block).

Substance use included any use of alcohol,tobacco, and illicit drugs; we did not includesmoking separately because of low prevalence(< 10%). We created mother’s psychologicalresources from 4 baseline measures: IQ,22

mental health,23 mastery,24 and self-efficacyrelated to behavioral objectives of NFP follow-ing Bandura’s model.25 We captured healthbeliefs through timing of prenatal care initia-tion, timing of program enrollment, andchild-rearing attitudes (higher scores on theAdolescent---Adult Parenting Index indicatehigher risk of child maltreatment26).

We selected 5 outcomes affected by theintervention in the Memphis trial and otherNFP trials.11,20,21,27---31We measured 2 out-comes near the child’s second birthday: (1)score on the Home Observation for Measure-ment of the Environment,32 and (2) mother’ssubsequent pregnancy. We collected 3 childoutcomes near age 12 years. We measuredinternalizing disorders with the Youth Self-Report33 and defined them as “overcontrol ofemotions—includ[ing] social withdrawal, de-mand for attention, feelings of worthlessness orinferiority, and dependency.”34(p677) Followingthe original study, we dichotomized internaliz-ing disorders to indicate clinical or borderlinedisorder.21 Reading and math achievementwere measured using the Peabody Individ-ual Achievement Tests.35 We limited ourachievement score analyses to children ofmothers with low psychological resources,because the intervention affected only thisgroup.19,36

Statistical Analysis

We employed repeated measures latentclass analysis. We used PROC LATENTCLASS ANALYSIS in SAS version 9.3 (SASInstitute, Cary, NC)37 to identify visit attendanceclasses (groups of mothers) on the basis ofnurse visit patterns. Latent class analysis isused to identify homogeneous subsamples of

RESEARCH AND PRACTICE

e2 | Research and Practice | Peer Reviewed | Holland et al. American Journal of Public Health | Published online ahead of print August 14, 2014

respondents from a population with a mixturedistribution.38,39 Repeated measures latentclass analysis is a nonparametric approachgrounded on categorical indicators of classmembership that can model patterns of changeover time.38 We chose this approach becausemethods with continuous measures requireassumptions of a functional form of patterns(such as polynomial), which were not likely tomodel visit patterns well.

We created periods so that the recommen-ded frequency of visits was constant withineach period (Figure 1). Because there wassubstantial variation in visit attendance duringthe first 6 postnatal weeks, we separated thistime into 2 periods (1---3 weeks and 4---6weeks). We calculated the percentage of visitscompleted compared with recommended foreach period and created a dichotomous vari-able indicating whether the mother attendedmore than 50% of the recommended visits. Wedetermined the number of classes from modelfit statistics (Akaike information criterion40 andadjusted Bayesian information criterion41) aswell as interpretability of patterns.

Because race is associated with attrition,3 weexamined models including only Black mothers

(90%), tested for measurement invariance byrace, and found no substantial differences invisit patterns. To maximize power, we havereported results that include all races.

We used bivariate analysis to identify vari-ables associated with class membership. Falsediscovery rate control is used in multiplehypothesis testing to correct for multiple com-parisons. Because of the large number ofmaternal characteristics included in the bivar-iate analyses, to control the expected propor-tion of incorrectly rejected null hypotheses(“false discoveries”), we used the false discov-ery rate method to determine the appropriatecutoff for statistical significance, for an overallfalse-positive rate of 5%.42 Multiple regressionanalyses included predictors identified fromthe bivariate analysis, with P< .10. In cases ofsignificantly correlated variables (P < .05), weselected the variable with the lowest P value.

We examined dichotomous outcomes (sub-sequent pregnancies, internalizing disorders)using logistic regression and continuous out-comes (Home Observation for Measurement ofthe Environment, math and reading achieve-ment) using linear regression. Following theoriginal study, we controlled for neighborhood

poverty, mother’s child-rearing attitudes, andpsychological resources.21

We compared outcomes for each interven-tion group class to the full control group and tomatched comparison groups constructedwithin the control group for each class. Wegenerated class assignment in the control groupby predicting each control mother’s likely classon the basis of her intake characteristics,assuming mothers with similar intake charac-teristics would follow similar visit attendancepatterns if they had been given the opportunity.We used multiple imputation with chainedequations43 using 100 imputations44 and in-cluded class membership as a missing variablefor each control group mother.

Note that class membership is missing com-pletely at random, because trial randomizationdetermined treatment group assignment. Asupplement to this article at http://www.ajph.org contains the variables we used for theimputation model, along with comparisonsbetween the intervention and control groupson these variables. For each class, we comparedthe intervention group outcomes with thecorresponding matched control groupoutcomes.

Most intake variables had few missing values(£ 2%). Higher rates of missingness were presentfor father involved (6%), parenting attitudes(11%), neighborhood poverty (11%), and racialconcordance with the nurse (12%). The 12-yearoutcomes were missing for 16% to 17%.

The Home Observation for Measurement ofthe Environment was missing for 6%, andsubsequent pregnancies were missing for 3%.We did not use imputation for missing databecause the overall rate of missing was low andimputation is not appropriate for outcomevariables.

On the basis of our findings, we conductedseveral analyses to further understand visitattendance classes. We examined birth out-comes, the mother’s school and employmentafter the child’s birth, and the nurse---motherrelationship. We used a Nurse---Client Rela-tionship Inventory, which independent inter-viewers conducted near the child’s secondbirthday and which consisted of 27 items witha 5-level Likert-scale describing the mother’srelationship with her nurse (Cronbach a=0.96). Lower values correspond to betterrelationships.

60

80

100

120

140

160

Ave

rage

Rec

omm

ende

d Vi

sits

Att

ende

d, %

High

Increasing

Low

0

20

40

-4 4 8 12 16 20 24Months(birth)

0

Note. The midpoint of each period represents that period on the graph. Period 1 = 1–4 weeks following enrollment;

recommended visit frequency = weekly. Period 2 = 5 weeks after enrollment until birth; recommended visit frequency = every 2

weeks. Period 3 = birth through 3 weeks; recommended visit frequency = weekly. Period 4 = 4–6 weeks; recommended visit

frequency = weekly. Period 5 = 7 weeks–12 months; recommended visit frequency = every 2 weeks. Period 6 = 12–20 months;

recommended visit frequency = every 2 weeks. Period 7 = 21–24 months; recommended visit frequency = monthly.

FIGURE 1—Visit completion trajectories of each class by periods: Nurse–Family Partnership;

Memphis, TN; June 1990–March 1994.

RESEARCH AND PRACTICE

Published online ahead of print August 14, 2014 | American Journal of Public Health Holland et al. | Peer Reviewed | Research and Practice | e3

RESULTS

Both 3- and 4-class models fit the data. Wechose the 3-class model on the basis of thesmall difference between the 3- and 4-classmodel for both Akaike information criterionand adjusted Bayesian information criterion(Table 1) and the simpler interpretation offewer classes. Figure 1 illustrates these classes.“High attenders” (48%) attended at least 50%of recommended visits throughout the pro-gram. “Increasing attenders” (18%) increasedtheir average visit attendance from 43% justafter birth to 96% in the second year. “Lowattenders” (33%) attended fewer than 50% ofrecommended visits in most periods. Highattenders had more visits throughout the pro-gram (50.1 visits of 61.5 recommended) thandid increasing attenders (42.4; P< .003) andlow attenders (22.9; P< .001).

The only statistically significant bivariateassociation between class membership andpredictors was highest grade completed inschool (Table 2). Low attenders completed themost grades in school, and increasing attenderscompleted the fewest. In multivariable models(Table 3), high and increasing attendance wereboth associated with less education than waslow attendance.

For the intervention group, we compared out-comes for each class with each other and detectedno significant differences (Table 4; data availableas a supplement to this article at http://www.ajph.org). High and low attenders had better homeenvironments than did the full control group(Table 4). High attenders had fewer subsequentpregnancies than did the full control group. Wedid not see any differences between any of the

classes and the full control group for schoolachievement or internalizing disorder outcomes.

Evidence of successful control groupmatching is presented in a supplement tothis article at http://www.ajph.org. The onlydifferences between the intervention andmatched control groups on intake characteris-tics were that increasing attenders had lowereducation than did their matched controls andhigh attenders talked to their baby’s father lessoften than did their matched controls. Highattenders had better home environments andchildren with higher math achievement thandid their matched control group. Low attendershad fewer subsequent pregnancies. No differ-ences were seen between any of the classes andmatched control groups for the child’s readingachievement or internalizing disorders.

Increasing attenders had greater likelihoodof preterm birth (18%) than did high attenders(4%; P< .005) but not low attenders (8%;P< .24). Babies of increasing attenders hadlonger hospital stays (11 days; P< .001) thandid those of high attenders (3.7) and lowattenders (3.7). Increasing attenders had in-fants of lower gestational age (38.0 weeks;P< .002) than did high (39.7) and low (39.4)attenders. Birth weight (P< .051) and Apgarscores (P< .31) were not associated with classmembership (means shown in a supplement tothis article at http://www.ajph.org).

Among increasing attenders, the infants ofmatched controls had lower 5-minute Apgarscores and younger gestational age at birththan did intervention babies (data available asa supplement to this article at http://www.ajph.org). Among low attenders, the infants ofmatched controls also had younger gestationalage than did intervention babies.

School attendance at 12 and 24 months,employment at 12 and 24 months, and sub-sequent pregnancy by 12 months were notassociated with class membership (data avail-able as a supplement to the online version ofthis article at http://www.ajph.org). High andincreasing attenders reported better relation-ships with their nurses (1.74 and 1.73, respec-tively) than did low attenders (2.04; P< .002).

DISCUSSION

The 3 visit attendance patterns were asso-ciated with predictors and outcomes. High

attenders constituted the majority of themothers in this study. Their intake character-istics and outcomes fell between the other 2groups. This group most closely followed therecommended visit schedule, and therefore wehave used it as the reference group in ourdiscussion.

Low attenders had the highest education.Although no other characteristics reachedstatistical significance after false discoveryrate adjustment, from an exploratory perspec-tive other characteristics suggest that thesemothers may be better situated than arethe other groups (they are less likely to bein the youngest age category and they havemore psychological resources at enrollment).Low attending mothers may have had lessneed for the program, and they and theirnurses likely recognized this and, as they hadbeen guided to do, adjusted the visitationschedule to reflect the fewer needs of the lowattenders.

This pattern of results is consistent with anearlier analysis of total visits completed, whichfound that mothers with moderately highpsychological resources had the fewest visits,whereas mothers with very high or very lowpsychological resources had more visits.12 Theoutcomes for these mothers and their childrenwere at least as good as those of high attenders,and they had fewer subsequent pregnanciesthan did their matched control group. Thissuggests that these families reaped the benefitsfrom the program in the early months whilethey were attending, and a lower level ofsubsequent involvement at least was notharmful.

Increasing attenders, by contrast, had in-take characteristics that were generallypoorer than were those of the other groups.They had significantly less education andhigher rates of preterm delivery and neonatalintensive care use. Unadjusted significancetests suggest that they may have lower psy-chological resources and may experiencegreater neighborhood poverty. Although thisgroup received the second highest number ofvisits, their outcomes were worse than werethose of the other groups. Children in thisgroup were likely to have been perceivedas being more vulnerable, which probablymotivated both the nurse and the motherto complete visits after delivery. A longer

TABLE 1—Model Fit Statistics for

Repeated Measures Latent Class

Analysis: Nurse–Family Partnership;

Memphis, TN; June 1990–March 1994

No. of Classes AIC Adjusted BIC Entropy

1 261.43 262.75 1.00

2 151.85 154.67 0.70

3 138.75 143.07 0.70

4 138.22 144.05 0.71

Note. AIC = Akaike information criterion; BIC = Bayes-ian information criterion.

RESEARCH AND PRACTICE

e4 | Research and Practice | Peer Reviewed | Holland et al. American Journal of Public Health | Published online ahead of print August 14, 2014

hospital stay is likely to have delayed receiptof postpartum visits.

The low attendance pattern may reflect thenurses’ choosing to focus their limited time onthose with premature babies, who were clearlyin greater need. Although we were not able todetermine how individual nurses managedtheir caseloads, we think that the risks of the

increasing attender group were greater thanthose of the low attender group, and the nurses’corresponding calculation of need likely ac-counts for the shift in patterns of visitationbetween these 2 groups. It is revealing, wethink, that mothers in the increasing attendergroup reported better relationships with theirnurses than did low attenders, whose needs

and frequency of visits were lower, especiallyafter delivery.

Implications

The identified classes suggest that a simpledose---response relationship does not bestdescribe visit attendance in this program.Future studies examining visit attendance

TABLE 2—Bivariate Associations Between Class Membership and Mother’s Intake Characteristics: Nurse–Family Partnership; Memphis, TN;

June 1990–March 1994

Variable

Class 1: High Attenders,a %,

Mean (SD), or Mean (Range)

Class 2: Increasing Attenders,a

%, Mean (SD), or Mean (Range)

Class 3: Low Attenders,a %,

Mean (SD), or Mean (Range)

Difference Between

Classes, Pb

Program

Racial concordance with first nurse 36.3 41.0 42.9 .26

Nurse changed during program 52.5 (50.1) 71.1 54.9 .02

Demographics

Aged < 17 y 43.1 (13–16) 38.5 (13–16) 19.7 (15–16) .005

Aged > 18 y 34.3 (19–29) 33.3 (19–33) 43.7 (19–26) .78

Mother’s race is White 8.8 7.7 14.1 .45

Married 2.0 0.0 1.4 .57

Lives alone 1.0 0.0 2.9 .33

In school 64.7 53.8 46.5 .08

Highest completed grade in school 9.9 (2.1) 9.4 (1.9) 10.9 (1.7) £ .001*Employed at enrollment 2.9 12.8 8.5 .13

Father involvedc 84.4 91.7 91.0 .1

Received government assistanced 83.1 83.3 79.3 .42

Household discretionary income –543.0 (6718.0) –397.0 (6398.0) 1270.0 (6889.0) .14

Neighborhood povertye 35.8 (20.5) 40.2 (21.2) 33.7 (18.5) .05

Substance use during pregnancyf 10.8 15.4 14.1 .78

Psychological characteristics

Mental healthg 99.7 (10.6) 97.5 (8.7) 99.6 (11.7) .18

Masteryh 98.2 (10.2) 99.7 (10) 101.5 (9.5) .43

Personal psychological resourcesi 98.7 (11.3) 97 (9.9) 102.2 (10.2) .02

Health beliefs

Prenatal care, time of first health care visit

in wk after last menstrual period

15.4 (6.3) 15.6 (5.2) 17.4 (6.6) .06

Parenting attitudesj 102.2 (8.8) 101.3 (8.5) 99.4 (8.2) .06

Estimated gestation at enrollment, wk 16.5 (5.9) 16.5 (5.5) 17.5 (5.5) .46

Note. Age reference = 17–18 years.aHigh attenders attended at least 50% of recommended visits throughout the program. Increasing attenders had low visit attendance early in the program but increased attendance during the firstyear of the child’s life. Low attenders had high visit attendance before birth but decreased attendance after birth.bLog-likelihood test for significance.cMother had contact with the child’s father.dReceived assistance from government programs, including public assistance, food stamps, Medicaid, and Social Security.ePercentage of households at or below federal poverty level (based on the U.S. Census Bureau definition).fSelf-reported substance use, including drugs, alcohol, and tobacco.gRAND Mental Health Inventory.hPearlin Mastery Scale. Mastery is the extent to which the mother believes she can control her own life outcomes.iPersonal psychological resources on the basis of intelligence (Shipley), mastery, mental health, and efficacy; dichotomized by sample mean for categorical variable.jAdult-Adolescent Parenting Inventory. Higher scores are associated with a higher likelihood of child abuse and neglect.*P < .05 after false discovery rate correction (cutoff = 0.0024).

RESEARCH AND PRACTICE

Published online ahead of print August 14, 2014 | American Journal of Public Health Holland et al. | Peer Reviewed | Research and Practice | e5

should consider patterns of attendance inaddition to dose.

Highest grade completed may be a goodpredictor of attendance patterns because many

factors contribute to school success, such asaptitude, stable environment, supportive fam-ily, and age. These factors may also be relatedto a mother’s visit attendance. Although we did

not find any other characteristics to besignificant, a larger sample size and differentmeasures may provide additional valuablepredictors.

TABLE 3—Multinomial Logistic Model Predicting Class Membership From Intake Characteristics: Nurse–Family Partnership; Memphis, TN; June

1990–March 1994

Characteristic

Odds of Being a High vs Increasing

Attender,a OR (95% CI)

Odds of Being an Increasing vs Low

Attender, OR (95% CI)

Odds of Being a High vs Low

Attender, OR (95% CI)

Highest grade completed 1.33 (0.91, 1.94) 0.59* (0.41, 0.84) 0.78* (0.63, 0.97)

Nurse changed during program 0.33 (0.07, 1.46) 2.30 (0.55, 9.60) 0.76 (0.34, 1.72)

First prenatal visit, wk after last menstrual period 0.98 (0.98, 1.01) 0.99 (0.98, 1.01) 0.99 (0.98, 1.002)

Father involvedb 0.13 (0.004, 3.91) 2.64 (0.07, 97.40) 0.35 (0.093, 1.30)

Note. CI = confidence interval; OR = odds ratio. The final model contains variables with uncorrected P < .05 in bivariate analyses. We eliminated age, personal psychological resources, school,parenting attitudes, and neighborhood poverty because of significant correlation with highest grade completed.aHigh attenders attended at least 50% of recommended visits throughout the program. Increasing attenders had low visit attendance early in the program but increased attendance during the firstyear of the child’s life. Low attenders had high visit attendance before birth but decreased attendance after birth.bMother had contact with the child’s father.*P < .05.

TABLE 4—Associations Between Class Membership and Outcomes: Nurse–Family Partnership; Memphis, TN; June 1990–April 1994,

July 2003–December 2006

Outcome (When Data Collected) High Attenders,a % or Mean (SE) Increasing Attenders,a % or Mean (SE) Low Attenders,a % or Mean (SE) Full Control Group, % or Mean (SE)

Subsequent pregnancy (24 mo)

Intervention group 41*b 41 28**c

Matched control group subsampled 47 48 47 48

Home environmente (24 mo)

Intervention group 31.7*b,c (5.7) 31.2 (6.8) 32.8**b (5.2)

Matched control group subsampled 30.7 (5.7) 30.2 (5.9) 31.9 (5.5) 31.0 (5.7)

Math achievementf (12 y)

Intervention group 87.3*c (11.7) 83.6 (9.1) 89.9 (9.6)

Matched control group subsampled 86.0 (9.9) 85.9 (10.3) 88.4 (10.5) 86.8 (10.3)

Reading achievementf (12 y)

Intervention group 88.4 (11.6) 86.4 (10.5) 91.6 (11.8)

Matched control group subsampled 89.4 (12.2) 88.7 (12.7) 91.1 (11.2) 89.8 (12.1)

Internalizing problem behaviorsg (12 y)

Intervention group 26 20 27

Matched control group subsampled 31 31 31 32

Note. Intervention group (n = 212) had 48% in high attendance, 18% in increasing attendance, and 33% in low attendance. The control group (n = 514) had imputed class membership of 39% inhigh attendance, 27% in increasing attendance, and 34% in low attendance. We tested outcomes between classes for the intervention group only and did not find any significant associations.aHigh attenders attended at least 50% of recommended visits throughout the program. Increasing attenders had low visit attendance early in the program but increased attendance during the firstyear of the child’s life. Low attenders had high attendance before birth but decreased attendance after birth.bContrast of each class to the full control group.cContrast of each class to the matched control group.dWe used multiple imputation with chained equations to create matched subsamples of the control group on the basis of intake variables.eHome Observation for Measurement of the Environment. Higher scores indicate a home environment that better supports child development.fPeabody Individual Achievement Test at 12 years, low-resource only. (We defined low-resource mothers as the bottom 50% of the sample for an aggregate measure of baseline IQ, mastery [theextent to which the mother believes she can control her own life outcomes, and parenting self-efficacy.) Higher scores indicate greater achievement. Intervention group (n = 113) had 52% in highattendance, 20% in increasing attendance, and 27% in low attendance. The average control group per imputation (n = 256) had imputed class membership of 42% in high attendance, 32% inincreasing attendance, and 27% in low attendance.gYouth Self-Report (Achenbach), percentage of children with scores above the borderline clinical threshold.*P < .05; **P < .01.

RESEARCH AND PRACTICE

e6 | Research and Practice | Peer Reviewed | Holland et al. American Journal of Public Health | Published online ahead of print August 14, 2014

Prediction of class membership duringpregnancy and the early weeks following de-livery may be a useful approach to managingnurse caseloads. A nurse who has many clientswho miss appointments during pregnancy andeventually become increasing attenders mayhave more difficulty providing needed care toher families than does a nurse with low-riskclients, who may actually need fewer visits afterdelivery.

Caseload management is important not onlyto provide adequate care to families but alsoto prevent burnout and turnover amongnurses. Participant attrition increases nurseturnover,3 along with hiring and training costs.Olds and his group are developing a tool fornurses to use that more comprehensivelycharacterizes families’ strengths and risks (theStrength and Risk framework), which may helpnurses understand families’ needs for the pro-gram and make adjustments in visit frequencythat align with families’ perceptions of theirneeds.45 Our findings provide empirical evi-dence to support use of this framework.

Limitations

The data we examined were from a single,closely supervised research trial site. DifferentNFP sites with fewer resources for supervisionmay find different visit attendance patterns.The specific patterns we found therefore maynot reflect ongoing NFP practice. Moreover, themoderate sample size in this study led to classeswith few participants, which limited statisticalpower. This is an issue that was particularlychallenging for the analysis of achievement testoutcomes, which employed only half of thesample.

It also is important to keep in mind thatthe data for this study came from a trialconducted in the early 1990s. Changes intechnologies (such as cell phones) and socio-cultural changes may also affect visit atten-dance in today’s environment. Replication ofthese analyses with currently operating NFPprograms and with different programs wouldprovide valuable information about visitdosage patterns and families’ needs andoutcomes.

Conclusions

Studying visit pattern trajectories is likelyto help us understand visit dosage, leading

to better allocation of scarce visitor time.The exploration of interactions betweennurses and mothers with respect to schedul-ing and attending visits is likely to improvehome visiting programs in communitypractice. j

About the AuthorsAt the time of this study, Margaret L. Holland was with theSchool of Nursing, Yale University, West Haven, CT.Yinglin Xia was with the Department of Biostatistics andComputational Biology, University of Rochester, Rochester,NY. Harriet J. Kitzman was with the School of Nursing,University of Rochester. Ann M. Dozier was with theDepartment of Public Health Sciences, University ofRochester. David L. Olds was with the Department ofPediatrics, University of Colorado, Aurora.Correspondence should be sent to Margaret L. Holland,

School of Nursing, Yale University, PO Box 27399, WestHaven, CT 06516-7399 (e-mail: [email protected]). Reprints can be ordered at http://www.ajph.org byclicking the “Reprints” link.This article was accepted May 28, 2014.

ContributorsM. L. Holland originated and designed the study, ana-lyzed the data, and led the writing of the article. Y. Xiasupported data analysis and contributed to the article.H. J. Kitzman contributed to the conceptualization of thestudy. H. J. Kitzman and D. L. Olds supplied the data. H. J.Kitzman, A. M. Dozier, and D. L. Olds contributed tointerpretation of the results and revision of the article.

AcknowledgmentsThis project was supported by the University of Roches-ter, Clinical & Translational Science Institute, the NationalCenter for Research Resources, and the National Centerfor Advancing Translational Sciences of the NationalInstitutes of Health (award KL2 RR024136). The Uni-versity of Rochester School of Nursing Faculty ResearchSeed Grant provided additional support.

D. L. Olds helped develop the Nurse Family Partner-ship and was principal investigator on the original trials ofthe program. H. J. Kitzman was co-principal investigatoron the Memphis trial and led the intervention and itsimplementation and fidelity.

Note. The content is solely the responsibility of theauthors and does not necessarily represent the officialviews of the National Institutes of Health.

Human Participant ProtectionThe Yale University human subjects committee deter-mined this study to be exempt because it involved onlyexisting, de-identified data.

References1. Dumas JE, Lynch AM, Laughlin JE, Smith EP, PrinzRJ. Promoting intervention fidelity—conceptual issues,methods, and preliminary results from the EARLYALLIANCE prevention trial. Am J Prev Med. 2001;20(1):38---47.

2. Ingoldsby EM, Baca P, McClatchey MW, et al. Quasi-experimental pilot study of intervention to increaseparticipant retention and completed home visits in the

Nurse---Family Partnership. Prev Sci. 2013;14(6):525---534. doi: 10.1007/s11121-013-0410-x.

3. O’Brien RA, Moritz P, Luckey DW, McClatchey MW,Ingoldsby EM, Olds DL. Mixed methods analysis ofparticipant attrition in the Nurse---Family Partnership.Prev Sci. 2012;13(3):219---228.

4. Raikes H, Green BL, Atwater J, Kisker E, ConstantineJ, Chazan-Cohen R. Involvement in Early Head Starthome visiting services: demographic predictors and re-lations to child and parent outcomes. Early Child Res Q.2006;21(1):2---24.

5. Wagner M, Spiker D, Gerlach-Downie S, HernandezF. Parental Engagement in Home Visiting Programs—Findings From the Parents as Teachers Multisite Evalua-tion. Menlo Park, CA: SRI International; 2000.

6. Korfmacher J, Kitzman H, Olds D. Intervention pro-cesses as predictors of outcomes in a preventive home-visitation program. J Community Psychol. 1998;26(1):49---64.

7. Ammerman RT, Stevens J, Putnam FW, et al. Pre-dictors of early engagement in home visitation. J FamViolence. 2006;21(2):105---115.

8. Coatsworth JD, Duncan LG, Pantin H, Szapocznik J.Patterns of retention in a preventive intervention with ethnicminority families. J Prim Prev. 2006;27(2):171---193.

9. Duggan A, Windham A, McFarlane E, et al. Hawaii’sHealthy Start Program of home visiting for at-riskfamilies: evaluation of family identification, family en-gagement, and service delivery. Pediatrics. 2000;105(suppl 2):S250---S259.

10. Whittaker KA, Cowley S. An effective programme isnot enough: a review of factors associated with poorattendance and engagement with parenting supportprogrammes. Child Soc. 2010;26(2):138---149.

11. Kitzman H, Olds DL, Henderson CR Jr, et al. Effectof prenatal and infancy home visitation by nurses onpregnancy outcomes, childhood injuries, and repeatedchildbearing. A randomized controlled trial. JAMA.1997;278(8):644---652.

12. Olds DL, Korfmacher J. Maternal psychologicalcharacteristics as influences on home visitation contact.J Community Psychol. 1998;26(1):23---36.

13. Holland ML, Christensen JJ, Shone LP, Kearney MH,Kitzman HJ. Women’s reasons for attrition from a nursehome visiting program. J Obstet Gynecol Neonatal Nurs.2014;43(1):61---70.

14. Baker CN, Arnold DH, Meagher S. Enrollment andattendance in a parent training prevention program forconduct problems. Prev Sci. 2011;12(2):126---138.

15. Haskins R, Paxson C, Brooks-Gunn J. Social ScienceRising: A Tale of Evidence Shaping Public Policy. Prince-ton, NJ: Future of Children; 2009.

16. Olds DL, Henderson CR Jr, Tatelbaum R,Chamberlin R. Improving the delivery of prenatal careand outcomes of pregnancy: a randomized trial of nursehome visitation. Pediatrics. 1986;77(1):16---28.

17. Olds DL. Prenatal and infancy home visiting bynurses: from randomized trials to community replication.Prev Sci. 2002;3(3):153---172.

18. Olds DL, Henderson CR Jr, Kitzman HJ, EckenrodeJJ, Cole RE, Tatelbaum RC. Prenatal and infancy homevisitation by nurses: recent findings. Future Child. 1999;9(1):44---65, 190---191.

19. Olds DL, Robinson J, Pettitt L, et al. Effects of homevisits by paraprofessionals and by nurses: age 4 follow-up

RESEARCH AND PRACTICE

Published online ahead of print August 14, 2014 | American Journal of Public Health Holland et al. | Peer Reviewed | Research and Practice | e7

results of a randomized trial. Pediatrics. 2004;114(6):1560---1568.

20. Olds DL, Kitzman H, Hanks C, et al. Effects of nursehome visiting on maternal and child functioning: age-9follow-up of a randomized trial. Pediatrics. 2007;120(4):e832---e845.

21. Kitzman HJ, Olds DL, Cole RE, et al. Enduringeffects of prenatal and infancy home visiting by nurses onchildren: follow-up of a randomized trial among childrenat age 12 years. Arch Pediatr Adolesc Med. 2010;164(5):412---418.

22. Shipley W. A self-administering scale for measuringintellectual impairment and deterioration. J Psychol.1940;9(2):371---377.

23. Ware JE, Veit CT, Donald CA. Refinements in theMeasurement of Mental Health for Adults in the HealthInsurance Study. Santa Monica, CA: RAND; 1985.

24. Pearlin LI, Schooler C. The structure of coping.J Health Soc Behav. 1978;19:2---21.

25. Bandura A. Self-efficacy: toward a unifying theory ofbehavioral change. Psychol Rev. 1977;84(2):191---215.

26. Bavolek SJ. Handbook for the AAPI: Adult-AdolescentParenting Inventory. Eau Claire, WI: Family DevelopmentResources, Inc.; 1984.

27. Kitzman H, Olds DL, Sidora K, et al. Enduring effectsof nurse home visitation on maternal life course: a 3-yearfollow-up of a randomized trial. JAMA. 2000;283(15):1983---1989.

28. Olds D, Henderson CR Jr, Cole R, et al. Long-termeffects of nurse home visitation on children’s criminal andantisocial behavior: 15-year follow-up of a randomizedcontrolled trial. JAMA. 1998;280(14):1238---1244.

29. Olds DL, Eckenrode J, Henderson CR Jr, et al. Long-term effects of home visitation on maternal life courseand child abuse and neglect. Fifteen-year follow-up ofa randomized trial. JAMA. 1997;278(8):637---643.

30. Olds DL, Kitzman H, Cole R, et al. Effects of nursehome-visiting on maternal life course and child devel-opment: age 6 follow-up results of a randomized trial.Pediatrics. 2004;114(6):1550---1559.

31. Olds DL, Kitzman HJ, Cole RE, et al. Enduringeffects of prenatal and infancy home visiting by nurses onmaternal life course and government spending: follow-upof a randomized trial among children at age 12 years.Arch Pediatr Adolesc Med. 2010;164(5):419---424.

32. Caldwell BM, Bradley RH. HomE Observation forMeasurement of the Environment. Little Rock: Universityof Arkansas, Center for Child Development and Educa-tion; 1979.

33. Achenbach TM, Rescorla LA. Manual for ASEBASchool-Age Forms & Profiles. Burlington, VT: University ofVermont, Research Center for Children, Youth, andFamilies; 2001.

34. Guttmannova K, Szanyi JM, Cali PW. Internalizingand externalizing behavior problem scores: cross-ethnicand longitudinal measurement invariance of the Behav-ior Problem Index. Educ Psychol Meas. 2008;68(4):676---694.

35. Markwardt FC. Peabody Individual AchievementTests---Revised. Circle Pines, MN: American GuidanceServices Inc.; 1998.

36. Olds DL, Robinson J, O’Brien R, et al. Home visitingby paraprofessionals and by nurses: a randomized, con-trolled trial. Pediatrics. 2002;110(3):486---496.

37. Collins LM, Lanza ST. Latent Class and LatentTransition Analysis With Applications in the Social,Behavioral, and Health Sciences. Hoboken, NJ: Wiley;2010.

38. Collins LM, Lanza ST. Latent Class and LatentTransition Analysis. Hoboken, NJ: Wiley; 2010.

39. Muthén BO. Latent variable hybrids: overview ofold and new models. In: Hancock GR, Samuelsen KM,eds. Advances in Latent Variable Mixture Models. Char-lotte, NC: Information Age Publishing; 2008:1---24.

40. Akaike H. Factor analysis and AIC. Psychometrika.1987;52(3):317---332.

41. Schwarz G. Estimating the dimension of a model.Ann Stat. 1978;6(2):461---464.

42. Xia Y, Chen R. False discovery rate. 2009. Availableat: ctspedia.org/do/view/CTSpedia/FalseDiscoveryRate.Accessed January 30, 2014.

43. Royston P. Multiple imputation of missing values:update of ice. Stata J. 2005;5(4):527---536.

44. White IR, Royston P, Wood AM. Multiple imputa-tion using chained equations: issues and guidance forpractice. Stat Med. 2011;30(4):377---399.

45. Olds D, Donelan-McCall N, O’Brien R, et al. Im-proving the nurse---family partnership in communitypractice. Pediatrics. 2013;132(suppl 2):S110---S117.

RESEARCH AND PRACTICE

e8 | Research and Practice | Peer Reviewed | Holland et al. American Journal of Public Health | Published online ahead of print August 14, 2014