AmerlcalljouNIal 011 Mellta/ Retardattoll199', Val. 97, No. ~, '59-'72e 199' American Ao.,)CI.tJon on Mental Retard.tJol1

Long- Term Outcome forChildren With Autism WhoReceived Early IntensiveBehavioral Treatment

JohnJ. McEachln, T..istram SmJth, and O.lvarLovaasUniv~rsity of California, Los Ang~l~s

After a very tntenstve behavtoral tnterventton, an e.xpeirtmental group oJr 19preschool-age chtldren with auttsm achteved less restrl(:;ttve school place~nents andhtgher IQs than dtd a control group of 19 stmtlar chtla'ren by age 7 (Lovaas,1987) .The present study followed-up thts ftnding by a.I;sesstng subjects at a meanage of 11.5 years. Results showed that the experimental group preseroed tts gainsover the control group. 11Je 9 experimental subjects whlo had achteved tl,e bestoutcomes at age 7 received parttcularly extenstve eval~~attons tndtcattn~~ that 8 ofthem were tndtsttngutshable from average chtldren on tests of tntelltgen,ce andadapttve behavtor. Thus. behavioral treatment may pn)duce long-Iasttn,g andstgntftcant gatns for many young chtldren wtth auttsm.

Infanttle auttsm is a conditionmarked by severe impairment in intellectual,social, and emotional functioning. Its onsetOCClKS in infancy, and the prognosis appears

to be extremely poor CLotter, 1978). Forexample, in the longest prospective follow-up study witJi1 a sound method,ological de-sign, Rutter (1970) found that only 1 of 64subjects with autism (fewer than 2%) couldbe considered free of clinical11' significantproblems by adulthood, as evidenced byholding a jo,b, living independently, andmaintaining :!n active and age..appropriatesodal life. 111e remaining subjt:cts showednumerous dysfunctions, such as markedoddities in behavior, social isolation, andflorid psychopathology .The majority of sub-jects required supervised living conditions.

Professionals have attempted a widevariety of interventions in an effort to helpchildren witl:l autism. For mar:ly years, noscientific evidence showed that any of theseinterventioru; brightened the children 's long-tenn progno:5is CDeMyer et al., 1981). How-

~\)

I This study was supported by Grant No. MH-11440 from the National Institute of Mental Health.The study was based on a dissertation submittedto the University of California, Los Angeles,Department of Psychology ' in partial fulfillmentof the requirements for the doctoral degree. Theauthors express their deep appreciation to themany students at UCLA who served as therapistsand helped to make this study possible. Specialthanks to Bruce Baker and Duane Buhrmester 1who helped in the design of this study. Requestsfor reprints of this article, copies of the ClinicalRating Scale, or additional information about thisstudy should be sent to 0. Ivar Lovaas, 405Hilgard Ave., UCLA, Department of Psychology,Los Angeles, CA 90024-1563.

ii

359McEachin, Smith, and lovaas

Also, as 'Was stated in the first follow-up(Lovaas. 1987). "Certain residual deficits mayremain in the normal-functioning group thatcannot be detected by teach,ers and parentsand can only be isolated on closer psycho-logical assessment, particularly as these chil-dren gro~v oldern (p. 8). This possibilitypoints to tJhe need for a more I:letailed assess-ment and for continued foUow-ups of thegroup OVf!r time.

11le present investigation contained twoparts: In dle first part we exaJ11ined whetherseveral years after the evaluation at age 7, theexperimental group in Lovaas's (1987) studyhad maint:a.ined its treatment gains. Subjectsin the experimental group and one of thecontrol groups completed sta'l1dardized testsof intellectual and adaptive functioning. Thegroups were then contrasted v..ith each other,and their current performance was com-pared to their performance cln previous as-sessments, 1lle second part of the investiga-tion focus:ed on those subjects who hadachieved tJhe best outcome at the end of firstgrade in the Lovaas (1987) st1L1dy (i.e.. the 9subjects w'ho were classified ~~s normal func-tioning OUlt of the 19 in tht! experimentalgroup). We examined the eJ[tent to whichthese best,-outcome subjects could be con-sidered fr(~e of autistic symptomatology .Atest battery was constructed to assess avariety of possible defidts: for example,idiosyncraltic thought pattern:). mannerisms.and inter~;ts; lack of close relationships withfamily and friends; difficulty iJ:l getting alongwith peoplle; relative weaknesses in certainareas of cognitive functioning, such as ab-stract reasoning; not working up to ability inschool; flatness of affect; ab~;ence or pecu-liarity in s(:nse of humor. Possible strengthsto be identified included norrnal intellectualfunctionin;g. good relationships with familymembers. :lbility to function independently.appropriate use of leisure time, and ad-equate so(:ialization with pel~rs. Numerousmethodological precautions were taken toensure ob}ectivity of the follo'w-up examina-tion.

ever, since the 19605, one of these interven-tions, behavioral treatment, has appearedpromising. Behavioral treatment has beenfound to increase adaptive beha viors such aslanguage and social skills, while decreasingdisruptive behaviors such as aggression(DeMyer , Hingtgen, &Jackson, 1981; Newsom& Rincover, 1989; Rutter, 1985). Further-more, behavioral treatment has been con-tinuously refined and improved as a result ofongoing research efforts at a number of sites(Lovaas & Smith, 1988).

Some recent evidence has indicatedthat behavioral treatment has developed tothe point that it can produce substantialimprovements in the overall functioning ofyoung children with autism (Simeonnson,Olley, & Rosenthal, 1987). Lovaas (1987)provided approximately 40 hours per weekof one-on-one behavioral treatment for aperiod of2 years or more to an experimentalgroup of 19 children with autism who wereunder 4 years of age. This intervention alsoincluded parent training and mainstreaminginto regular preschool environments. Whenre-evaluated at a mean age of 7 years, sub-jects in the experimental group had gainedan average of 20 IQ points and had mademajor advances in educational achievementNine of the 19 subjects completed first gradein regular (nonspecial education) classesentirely on their own and had IQs thatincreased to the average range. By contrast,two control groups totalling 40 children, alsodiagnosed as autistic and comparable to theexperimental group at intake, did not farenearly as well. Only one of the controlsubjects (2.5%) attained nom1al levels ofintellectual and educational functioning.

These data suggest that behavioral treat-ment is effective. However, the durability oftreatment gains is uncertain. In one priormajor study, Lovaas, Koegel, Simmons, andLong (1973) found that children with autismregressed following the termination of treat-ment. Other studies have shown that chil-dren with autism may display increased dif-ficulties when they enter adolescence(Kanner, 1971; Waterhouse & Fein, 1984).

360 Autism and Early Intervention

Method

Subjects and Background

Characteristics of the subjects and theirtreatment have been described elsewhere(Lovaas, 1987) and will only be summarizedhere. The initial treatment study contained38 children who, at the time of intake, werevery young (less than 40 months if mute, lessthan 46 months if echolalic) and had re-c~ived a diagnosis of autism from a licensedclinical psychologist or psychiatrist not in-volved in the study. These 38 subjects weredivided into an experimental group and acontrol group. The assignment to groupswas made on the basis of staff availability. Atthe beginning of each academic quarter,treatment teams were formed. The clinicdirector and staff members then determinedwhether any opening existed for intensivetreatment. If so, the next referral receivedwould enter the experimental group; other-wise, the subject entered the control group.The experimental group contained 19 chil-dren who received 40 or more hours perweek of one-to-one behavioral treatment for2 or more years. The control group wascomprised of 19 children who received amuch less intensive intervention (10 hours aweek or less of one-to-one behavioral treat-ment in addition to a variety of treatmentsprovided by community agencies, such asparent training or special education classes).The initial study also included a secondcontrol group, consisting of 21 children withautism who were followed over time by anearby agency but who were never referredfor this stlldy. However, these 21 subjectswere not available for the present investiga-tion. On standardized measures of intelli-gence, the second control group did notdiffer from either the experimental group orthe first control group at intake, nor did itdiffer from the first control group whenevaluated again when the subjects were 7years old. These findings suggest that, asmeasured by standardized tests, (a) the chil-dren with autism who were referred to us for

~~

treatment were comparable to dlildren withautism seen elsewhere and (b) the minimaltreatment pro'l'ided to the first control groupdid not alter intellectual functiolrling.

Statistical analysis of an extensive rangeof pretreatment measures confirI1rled that theexperimental group and control ,group werecomparable at intake and closely Irnatched onsuch important variables as IQ and severityof disturbancf:. The mean chronological age(CA) at diagnosis for subjects in the experi-mental group was 32 months. Thl~ir mean IQwas 53 (range' 30 to 82; all IQs are given asdeviation scores). The mean CA of subjectsin the control group was 35 months; theirmean IQ was'i6 (range 30to 80). Most of thesubjects were mute, all had gro:)S deficien-des in recepti1fe language, none: played withpeers or sho\lfed age-appropriate toy play,all were emotionally withdrawn, most hadsevere tantrurns, and all showed extensiveritualistic and stereotyped (self-stimulatory)behaviors. Thus, they appeared to be arepresentative' sample of childrf:n with au-tism (Lovaas, Smith, & McEachiJrl, 1989). Amore complel:e presentation of the intakedata was rep<J'rted by Lovaas (19187).

The childlren in the experim,ental groupand control gJroup received their respectivetreatments from trained studenlt therapistswho worked ilrl the child's home. 'llie parentsalso worked 'with their child, alrld they re-ceived extensive instruction and :;upervisionon appropriat4~ treatment techniques. when-ever possible. the children werf~ integratedinto regular preschools. The trf~atment Io-cused primal1ily on developing language,increasing SOI:ial behavior, and promotingcooperative pJlaywith peers alonl~ with inde-pendent and 'appropriate toy pl;ly, Concur-rently, substantial efforts were directed atdecreasing excessive rituals, tarltrums, andaggressive behavior. (For a m(]lre detaileddescription of the intervention plrogram, seethe treatment manual (Lovaas et a:l., 19801 andinstructional videotapes that supplement themanual (Lova:~ & Leaf, 19811.)

At the time of the pres en It follow-up(1984-1985), Ibe mean CA of the experimen-

~

McEachin, Smith, and Lovaas 361

tal group children was 13 years (range ~, 9 to ~,ithintake IQ or outcome IQ. ConsequeJ

19 years). A!I c?ildren who had achileved a.lthoughthetendencyforthefirstreferranormal functionIng by the ~ge of7 year~; had enter the experimental group created aended treatment by that point. (Normaljunc- t(~ntial bias, the data indicate that thisttontng was operationally defined as scoring unlikely;within the normal range on standardizedintelligence tests and successfully complet- Procedureing first grade in a regular, nonspedal edu-cation class entirely on one's own.) On the The assessment procedure incluother hand, some of the children who had a:;certaining school ]placement and adminot achieved normal functioning at 7 years of tf'ring three standardized tests. Informalage had, at the request of their parents, o:rl school placement was obtained fJremained in treatment. The length of time S\lbjects' parents, ~/ho classified then:that experimental subjects had been O'lt of b,eing in either a reB~lar or a special ed,treatment ranged from 0 to 12 years (mean = tilDn class (e.g. , a l:lass for children ,5), with the normal-functioning chili:lren a,ltism or mental retardation, languagehaving been out for 3 to 9 years (mean ,= 5). lays, multihandicaps, or learning disal

The mean age of S\lbjects in the control tic~). The three standardized tests wertgroup was 10 years (range 6 to 14). The fcillows:length of time that these children had been 1. Intelligence rest. The Wechsler 111out of treatment ranged from 0 to 9 years lil~enceScaleforChildren-Revised(Wecru(mean = 3). Thus, experimental subjects 1~~74) was administered when subjects ~tended to be older and had been ou.t of able to provide verbal responses. Thistreatment longer than had control subjects. cluded all 9 best-outl:ome experimental sThis difference in age occurred becausf~ the jects plus 8 of the ren1aining 10 experimelfirst referrals for the study were all assigned Sllbjects and 6 of the 19 control subjoots.to the experimental group due to the fact that Sllbjects who were nI)t able to provide Velreferrals came slowly (7 in the f1fSt 3.5 years) responses, the Leitt~r International Per:and therapists were available to treat all of fiance Scale (Leiter, 1959) and the Peabcthem. (As noted earlier, subjects were. as- Picture Vocabulary Tt~t-Revised(Dunn,ISsigned to the experimental group if th,era- were administered. All of these tests h.pists were available to treat them; otherwise, bt~en widely used for the assessmentthey entered the control group.) intellectual functionlng in children with

Statistical analyses were conducted to ti~;m (Short & Marcus, 1986).test whether a bias resulted from the ten- 2. 7be Vtne/and Adapttve Bebaldency for the f1fSt refeITals to go into the S(;a/es (Sparrow, Balla; & Cicchetti, 19texperimental group. For example, it is c:on- 111e Vineland is a structured interviewceivable that the first referrals could have ministered to parents assessing the extbeen higher functioning at intake or could to which their child exhibitS behaviors thave had a better prognosis than subseq'lent are !:leeded to cope effectively withreferrals. If so, the subject assignment proce- everyday environment.dure could have favored the experimental 3. The Pmona/tty Inventory for c,group. To assess this possibility, we corre- dJ'"en (Wirt, Lachar, Klinedinst, & Seat, 19ilated the order of referral with intake IQ and 111is measure is a 6c1O-item true-false quwith IQ at the first follow-up (age 7 years). tionnaire filled out by parents that assesPearson correlations were computed across the extent to which their children sh,both groups and within each group. These va.rious forms of PSJ'chological disturbaJanalyses indicated that the order in which (e.g., anxiety, depr~;sion, hyperactivity, aSllbjects were referred was not associated p~;ychotic behavior).

A,utism and Early Interven1362

I

1fiese three tests were intended to pro-vide a comprehensive evaluation of intellec-tual, social, and emotional functioning. All ofthe tests have been standardized on averagepopulations. Hence, they provide an objec-tive basis for comparing subjects to childrenwithout handicaps across the various areasthat they assess.

Data were obtained on all subjects ex-cept one girl in the control group, who wasknown to be institutionalized and function-ing very poorly. The 9 best-outcome subjects(those who had been classified as normalfunctioning at age 7) received particularlyextensive evaluations, as outlined later. Ofthe 28 remaining subjects, 17 were evaluatedby staff members in our treatment program,and 11 received evaluations from outsideagencies such as schools or psychologyclinics. (In some cases, the outside agenciesdid not administer all of the measures in thisbattery.)

Evaluation of Best-Outcome Subjec:ts.To ensure objectivity in the evaluation of thebest-outcome subjects, we arranged for blindadministration and scoring of all tests forthese subjects as follows. A psychologist notassociated with the study recruited advancedgraduate stlldents in clinical psychology toadminister the tests. The examiners were notfamiliar with the history of the children, andthe psychologist told them simply tha~ thetesting was part of a research study onassessment of children. The psychologistadvised them that the nature of the studynecessitated providing only certain standardbackground information: age, school place-ment and grade, and parent's name andphone number. To increase the heterogene-ity of the sample and to control for anyexaminer bias, each examiner also testedone or more S\.lbjects who were matched inage to the experimental subjects and had nohistory of behavioral disturbance. 'I11e exam-iners were randomly assigned an approxi-mately equal number of subjects for testingin the experimental group and the compari-son group. Two experimental subjects werenot living in the local area. Therefore, for

363McEachin, Smith, and lovaas

each of them" the psychologist recruited atester from th,e subject's hometown area aswell as an age-matched control sllbject, anddata were collected as just described. Inaddition, the child's examiner filled out aclinical rating scale following a structuredinterview that covered a list c,f standardtopic includling friendships, fa.mily rela-tions, and school and community activities.The interview was designed both for elicit-ing content al:ld for sampling inlterpersonalstyle. The ratil:lg scale consisted of 22 items,each scored O (best clinical status) to 3(marked devi:ince) points. 11le items weredesigned to il1lclude likely areas of difficultyfor children ~..ith autism of avel'age intelli-gence (e.g., compulsive or ritualiistic behav-ior, empathy for and interest iI:l others, asense of humor) as well as areas I()f potentialdifficulty for the general child population(e.g., depressled mood, anxiety, hyperactiv-ity). (11le complete scale and a copy ofinstructions for the clinical intenriew can beobtained by ~vriting to the third author),

ReslJllts

Experimental Versus Control Gl"OUp

This firs It section examines the overalleffects of tre~Ltment through comparison ofthe follow-up' data from the 19 S1Jbjects whorecetved the intensive (experimental) treat-ment to the d~lta from those who I~eceived theminimal (control) treatment. Data were ob-tained from alII subjects on school placementand from all but one subject in the controlgroup on IQ. On the Vineland, scores wereobtajned for 18 of 19 experimental subjectsand IS of IS' control subjects. 11le lowestavailability of follow-up scores was on thePersonality Inventory for d1ildrerl, with scoresfor IS experimental subjects and 12 controlsubjects,

The subjects in the control group whohad Personality Inventory forChlldren scoresdid not appf~ar to differ from sllbjects whowere missin~: these scores, as compared on

t tests for differences in intake IQ, IQ at 7years old, or IQ in the present study.

As noted earlier, 17 of the 29 subjectswho were not in the best-outcome groupwere evaluated by Project staff members, 11were evaluated by outside agencies, and 1was not evaluated. To check whether Projectstaff members were biased in their evalua-tions or in their selection of which subjectsto evaluate, we used t tests to comparesubjects they evaluated to those evaluatedby outside agencies on intake IQ, IQ at age7 years, and IQ in the present study. Nosignificant differences between subjectsevaluated by Project staff members and thoseevaluated by outside agencies were found.

School Placement. In the experimentalgroup, 1 of the 9 Sllbjects from the best-outcome group who had attended a regularclass at age 7 0. L.) was now in a specialeducation class. However, 1 of the other 10Sllbjects had gone from a special educationclass to a regtllar class and was enrolled in ajunior college at the time of this follow-up.The remaining experimental subjects hadnot changed their classification. Overall, then,the proportion of experimental subjects inregular classes did not change from the age7 evaluation (9 of 19, or 47%). In the controlgroup, none of the 19 children were in aregular class, as had been true at the age 7evaluation. The difference in classroom place-ment between the experimental group andthe control group was statistically significant,r (1, N= 38) = 19.05,p< .05.

Intellectual Functioni ng .The test Scoresfor the experimental group and control groupon intellectual functioning, adaptive andmaladaptive behaviors, and personality func-tioning are summarized in Table 1. As can beseen in the table, the experimental group atfollow-up had a significantly higher mean IQthan did the control group. This differencewas significant, 1(35) = 2.97, p < .01. Elevensubjects (58%) in the experimental groupobtained Full-Scale IQs of at least 8Oj only 3subjects (17%) in the control group did aswell. The scores were similar to those ob-tained by the experimental group and con-

trot group at age 7 (meanrespectively), .

of the current evaJuatil:>n.

Table 1

s'erimenlalM8i~n SDMeasure

Mean

84.5

5.1

73.1

75.5

71.16

10.'6

61.1~

4.0

32.4

28.426.926.8

26.88.2

10.23.9

laVinetand°

CommunicationDaily Uving S,killsSocializationAdaptive Behavior

CompositeMaladaptive Behavior

PICb !)calesMe;ln elevationScales > 70

"Vineland Adaptive Behavior Scale.for Children.

Adaptive and Ma'ladapttveOn the Vineland, the m.ean overallposite SCore was 72 in thegroup and 48 in the control group.

this test is 100,

tion, Daily Living, and Sodalizatioft

group consistently scored higher

in the control group, 1(31) = 2.39,p< .05.mean score for the control group was inclinically significant range whereasthe experimental group was not.

of clinically significant le'velsbehavior at ages 6 to 9 years;12 to 13 years; and 10 or above, atand older.) Thus, the firldings indicate

364

iors than did the control group.Pe~ona/ity Functiontng. Scores for the

experimental group and control group didnot differ on overall scale elevation, withmean tscores of62 and 65, respectively. (Onthis test, the mean t score for the generalpopulation is approximately 50 [SD = 101.) Tscores above 60 are considered indicative ofpossible or mild deviance, whereas t scoresabove 70 are viewed as suggesting a clini-cally significant problem, namely, one thatmay require professional attention. Therewas a significant interaction between thegroups and the individual scales on this test,F(15, 390) = 2.36, p < .01. Results of theTukey test indicated that the most reliabledifference between groups ocCllrred on thePsychosis scale, on which the experimentalsubjects had a mean of 78 and the controlsubjects had a mean of 104, F(1, 26) = 8.53,p < .01. Seven subjects in the experimentalgrou p scored in th e clinically preferred range(below 70), whereas no subjects in the con-trol group scored that low. Only one otherscale showed a significant difference, So-matic Concerns, F(1, 26) = 4.60,p< .05. Thecontrol subjects tended to display a belowaverage level of somatic complaints (mean of45 as compared to 54 for the experimentalsubjects).

Best-Outcome Versus NonclinicalComparison Group

A t test indicated no significant differ-ence in age between the best-outcorne groupand the comparison group of children with-out a history of clinically significant behav-ioral distllrbance. Subjects in the best-out-come group had a mean age of 12.42 years(range 10.0 to 16.25) versus 12.92 years(range 9.0 to 15.17) for the nonclinical com-parison group. Scores on the WISC-R andclinical rating scale were obtained for allsubjects; 1 experimental subject and 2nonclinical comparison subjects were miss-ing Vineland scores, and 2 experimentalsubjects and 1 non clinical comparison sub-ject were missing Personality Inventory for

:~65McEachin, Smith, and Lovaas

Children scores. Both the Vineland and Pler-sonality Inventory for Children ~"ere com-pleted by parents. In cases where the'sescores were not obtained, the p:lrents haddeclined to participate.

On the measures that providt~ standard-ized scores, the functioning of the best-outcome subjects was measured most pJre-cisely by comparing the best-outcome gro'Jpagainst the test norms. Therefore, this anaily-sis is of primary interest. Data for t]henonclinical comparison group are mainlyuseful in confi~img that the assessme:ntprocedures were valid and in pl:Oviding acontrast group for the one measu:re withQlutnonns, the Clinical Rating Scalt~. For thenonclinical comparison group, it ~I/ill sufficeto summarize the results as follov{s: On theWISC-R this group had mean I(~s of 116Verbal, 118 Performance, and 119 Pull-Scale.On the Vineland the group obtained meanstandard scores of 102 Communic:ation, 100Daily Living Skills, 102 Socialization, and 101Composite. The mean scale score on thePersonality Inventory for Children was '(9.1fius, the nonclinical comparison group dis-played above-average or average functiQln-ing across all areas that were asst~sed.

The next sectlion is focust'd on thefunctioning of the best-outcome group IonIQ, adaptive and maladaptive behavior, a'ndpersonality measures and contrasls the best-outcome subjects with the comparison SlIlb-jects on the Clinical Rating Scale.

Intellectual Functiontng. Table 2 pre-sents the IQ data for each subject in the best-outcome group and the mean scQ,res for thegroup. This table shows that, as a 'whole, the9 best-outcome subjects performc~d well onthe WISC-R. Their IQs placed th,em in thehigh end of the normal range, :~bout t~'NOthirds of an SD above the mean. Their Full-Scale IQs ranged from 99 to 136.

Subjects' score$ were evenly ldistribul:edacross a range from 80 to 125 on Verbal IQand from 88 to 138 Ion Performance IQ. Thesubjects averaged 31 points higher on perfor-mance IQ than Veft!)alJQ. Two of them 0. L.and A" G.) had at least a 20-point difference

Note. Infrm = Information, Simil = Similarities, Arith = Arithmetic, Vocab = Vocabulary , Compr = Compreheneion,PicC = Picture Completion, PicA = Picture Arrangement, BlkD = Block Design, ObjA = Object Assembly , Cod= Coding, VIO = Verbal la, Pia = Performance 10, and Full = Full-Scale la.

I

sonality Inventory for Children, as measuredby the three validity scales (lie, Frequency Iand Defensiveness). As can be seen from thetable, the subjew scored in the normal rangeacross all scales. They tended to score high-est on Intellectual-Screening, Psychosis, andFrequency. Intellectual-Screening assessesslow intellectual development, and psycho-sis and Frequency assess unusual or strangebehaviors. Only Intellectual-Screening wasabove the normal range, and this scale isaffected by subjects' early history. For ex-ample, the scale contains statements such as"My child first talked before he (she) was twoyears old," which would be false for the best-outcome subjew regardless of their currentlevel of functioning.

As Table 4 indicates, 4 best-outcomesubjew had a single scale elevated beyond

Table 3Scores on the Vlneland Adaptive Behavior Scalefor the Best-Outcome SubJects-

~aptive behavior Maladaptivebehavior~ect cam DlS Sac -Comp

R.S.M.C.M.M.L.B.J.L.D.E.A.G.B.W.B.R.Mean

83

119

119

107

77

93

101

83

98

10286

114112

948299

105

99

9298

10510888809883

94

616

24

1315

59

6.8

between Verbal and Perfonnance IQ.On each subtest of the WISC-R, the

mean for the general population is 10 (SD ~3). It can be seen from Table 2 that the best-outcome subjects scored highest on Similari-ties, Block Design, and Object Assembly.They scored lowest on Picture Arrangementand Arithmetic. Thus, the subjects consis-tently scored at or above average.

Adaptive and Maladaptive BehaVior.Table 3 presents the data for the best-out-come group on the Vineland Adaptive Be-havior Scales. It can be seen that the best-outcome group scored about average on theComposite Scale and on the sub scales forCommunication, Daily Living, and Socializa-tion. However, Table 3 shows that some ofthe best-outcome subjects had marginalscores, includingJ. L., B. W., andM. M. Evenso, all of the best-outcome subjects hadComposite scores within the normal range.

As can be seen in Table 3, on theMaladaptive Behavior Scale (Parts I and II),the mean score for the best-outcome groupindicated that, on average, these subjects didnot display clinically significant levels ofmaladaptive behavior. Three of them scoredin the clinically significant range versus onesubject in the non clinical comparison group,which had a mean of 7.7 on this scale.

Pe~onaltty Functioning. The results ofthe Personality Inventory for Children aresummarized in Table 4. The best-outcomesubjects obtained valid profiles on the Per-

Note. Com = Communication, DLS = Daily Living Skills, Soc= Socialization, Comp = Adaptive Behavior Composite.

989379

108103619774

92

Table 4T Scores on the Personality Inventory for Children for the Best-Outcorne Subjects

T score

treatment. In the present study we havereported data on these children at a mean ageof 13 years for subjects in the experimentalgroup and 10 years for those in the controlgroup. 11le data were obtained from a com-prehensive assessment battery.

11le main findings from the test batterywere as follows: First, subjects in the experi-mental group had maintained their level ofintellectual functioning between their previ-ous assessment at age 7 and the presentevaluation at a mean age of 13, as measuredby standardized intelligence tests.11leir meanIQ was about 30 points higher than that ofcontrol subjects. Second, experimental sub-jects also displayed significantly higher lev-els of functioning than did control subjectson measures of adaptive behavior and per-sonality. Third, in a particularly rigorousevaluation of the 9 subjects in the experi-mental group who had been classified asbest-outcome (normal-functioning) in theearlier study (Lovaas, 1987), the test resultsconsistently indicated that the subjects ex-hibited average intelligence and averagelevels of adaptive functioning. Some devi-ance from average was found on the person-ality test and tl1e clinical ratings. However ,this deviance appeared to derive from theextreme scores of one subject, J. L. (see Table2, 3, and 4). This subject also had beenremoved from nonspedal education classesand placed in a class for children withlanguage delays, and he obtained relatively

the clinically significant range and a 5th 0. L.)had nine scales elevated, including the high-est scores in the best-outcome group onIntellectual-Screening, psychosis, and Fre-quency. Thus, this subject appeared to ac-count for much of the elevation in scores onthese scales. By comparison, there were 3subjects in the nonclinical comparison groupwith at least one scale elevated.

Clinical Rattng Scale. On this scale, 8 ofthe best-outcome subjects scored between 0and 10, and the 9th 0. L.) scored 42. Themean was 8.8, with a standard deviation of12.9. The nonclinical comparison subjects allscored between 0 and 5 (mean = 1.7, SD =2.1). Because these SDs are unequal, weused a nonparametric statistic, a Mann-Whitney Utest, revealing a significantdiffer-~nce between groups, U= 19,p< .05. Thus,me best-outcome subjects displayed moredeviance than did the comparison subjects,but most of the deviance appeared to comefrom one subject, J. L.

~,

&1

~~

('i!i

~1':~ Ci

!~~1if;.

Discussion

" This study is a later and more extensivefollow-up of two groups of young subjectswith autism who were previously studied byLovaas (1987): (a) an experimental group (n;: 19) that had received very intensive behav-.oral treatment and (b) a control group (n =19) that had received minimal behavioral

367McEachin, Smith, and Lovaas

low scores (about 80) on the Verbal section into a treatment study is beyond the scop Iof the intelligence test and the Communica- the present report (see Kazdin, 1980; Kention section of the measure of adaptive & Norton-Ford, 1982; Spitz, 1986). Howe'behavior. Thus, he no longer appeared to be we note that we incorporated a large numnonnal-functioning. However, the reroain- of methodological safeguards in boiliing 8 subjects who had previously been original study (Lovaa:;, 1987) and the pre5classified as normal-functioning demonstrated investigation:average IQ, with intellectual perfonnance 1. The experimental ~:roup andevenly distributed across subtests, were able control group received equivalent asscto hold their own in regular classes, did not ment batteries at intal<e and were found tcshow signs of emotional disturbance, and very similar on a multitud{~ of importdemonstrated adequate developmentofadap- variables. Moreover, the number of conItive and sodal skills within the normal range. group subjects who w,ere predicted to achi.In addition, Sllbjective clinical impressions nclnnal functioning, had they received intof blind examiners did not discriminate them sive treatment, was approximately equalfrom children with no history of behavioral the number of experimental subjects ~disturbance. These 8 subjects ( 42% of the actually did achieve normal f\;lnctioning ~experimental group) may be judged to have int,ensive treatment (I.ovaas & Smith, 19Emade major and enduring gains and may be Thus, the subject assignment procedldescribed as "nonnal-functioning." By con- yielded groups that ~vere comparable pitrast, none of the control group subjects to treatment. This provided a, strong indiachieved such a favorable outcome, consis- tion that the superior functioning of Itent with the poor prognosis for children experimental group after treatment wa:with autism reported by other investigators result of the treatment itself rather thai(Preeman,Ritvo,Needleman,&Yokota,1985). biased procedure for assigning subjects

In order to evaluate this outcome, we the experimental group.must pay close attention to whether or not 2. All subjects rernainedin the grOUP!our methodology was sound. The adequacy which they were assigned at intake. Onl:of our methodology is crudal because the subjects dropped out, and they were Ioutcome in the present stlldy represents a replaced. Therefore, the ori~~inal corn pcmajor improvement over outcomes obtained tion of the groups W~j essentially preservcin previous experimental studies on the 3. All subjects w'ere independentlytreatment of children with autism (Rutter, agnosed as autistic by PhD or MD clinidal1985). The only reports of comparable out- and there was high agreement on the di~comes have come from uncontrolled case nosis between the independent clinidalstudies (e.g., Bettelheim, 1967), and subse- This provided evidence that subjects nquent investigations have indicated that these criteria for a diagnosis of autism.case studies grossly overestimated the out- 4. Prior to treatment, these subjecomes obtainable with the treatment that appeared to be comparable to those diawas provided. Similarly, reports of major nosed as having autism in lother researgains in other populations, such as large IQ investigations. Evidence for this comes frcincreases in children from impoverished the second control group that was incorpbackgrounds, also have been based on highly rated into the initial treatment study, 11questionable evidence (Kamin, 1974; Spitz, group was evaluated, by another resear,1986). Such reports have the potential to team (independent of ours), had similar I(cause a great deal of harm by misleading at intal<Je based on the sarole measuresconsumers and professionals, intelligence that we used, yet showed simil

A detailed description of all the meth- outcome data to those reported by othodological safeguards that should be built investigators. Additional evidence can I

368 Autism and Early Interventi

~

,,;!;0"', ,,.

0,..,"jll;"C

have been maintained for an ext'~nded pe-riod of time.

9. A wide range of measures ~ras admin-istered, avoiding overreliance on intelligencetests, which have limitations if use~d in iso:la-tion (e.g. , bias Iresulting from teaching to t:hetest, selecting :! test that would yield espe-dally favorable' results, failing to ~;sess othleraspects of functioning such as soclial compe-tence or school perfonnance) (Spitz, 19~16jZigler & Tricke~tt, 1978).

10. The u5e at follow-up of a nomlalcomparison group, standardized t(~sting, alrldblind rating allowed for an obj(~ctive, dle-tailed, and quantifiable assessment of tre:a.t-ment effectiveness. A particularl~, rigorousassessment wa:s given to those subjects wlhoshowed the most improvement.

Taken together, these safegtJards pro-vide considerable assurance that the favor-able outcome of the experiment:u subje,ctscan be attribul:ed to the treatment they Ire-ceived rather than to extraneous fa.ctors suchas improvement that would hav(: occurredregardless of treatment, biased proceduresfor selecting s1ubjects or assigning them togroups, or narrow or inappropri:lte assess-ment batteries.

Despite tlle numerous precal.ltions tllatwe have taken, several concernls may beraised about tlle validity of the r(:sults. Per-haps the most imp<l>rtant is that the assi~;n-ment to the experitnental or control groupwas made on the b~is of therapist availabil-ity rather than, a more arbitrary proceduresuch as alternating: referrals (assigning l:hefirst referral to the experimental group, I:hesecond to the contrpl group, the Ihird to t:heexperimental group, and so forth), Howev'er,it seems unlikely that the assignment V{asbiased in view of the pretreatment data wehave presented on the similarity b,~tween Itheexperimental and control grout:lS. On Itheother hand, we do not know as yet whetherthere exists a pretreatment variable that doespredict outcome but was not among the 19we chose, yet could have discrinlinated ibe-tween groups. In an earlier publication(Lovaas et al., 1989), we responded in some

derived from the similarity of our intake datato data reported by other investigato~ (Lovaaset al., 1989). For example, although Schoplerand his assodates (Schopler, Short, & Mesibov,1989) suggested that our sample had a highermean IQ than did other samples of childrenwith autism, their own data do not appear todiffer from ours (Lord & Schopler, 1989).111us, there is evidence that our subjectswere a typical group of preschool-age chil-dren with autism rather than a select groupofhigh-level children with autism who wouldhave been expected to achieve nonnal func-tioning with little or no treatment.

5. The first control group, which re-ceived up to 10 hou~ a week of one-to-onebehavioral treatment, did not differ at post-treatment from the second control group,which received no treatment from us. Bothgroups achieved substantially less favorableoutcomes than did the experimental group.Because all groups were similar at pretreat-ment, this result confirms that our subjectshad problems that responded only to inten-sive treatment rather than problems such asbeing noncompliant or holding back (mask-ipg an underlying, essentially average intel-lectual functioning that would respond tosmaller-scale interventions).

6. Subjects' families ranged from high tolow socioeconomic status, and, on average,they did not differ from the general popula-tion (Lovaas, 1987). 111us, although our treat-ment required extensive family participa-tion, a diverse group of families wasapparently able to meet this requirement.

7. The treatment has been described indetail (Lovaas et al., 1980; Lovaas & Leaf,1981), and the effectiveness of many compo-nents of the treatment has been demon-strated experimentally by a large number ofipvestigato~ over the past 30 years (cf.Newsom & Rincover I 1989). Hence, our treat-ment may be replicable, a point that isdiscussed in greater detail later .

8. The results of the present follow-up,which extended several yea~ beyond dis-charge from treatment for most subjects, arean encouraging sign that treatment gains

:369McEachin, Smith, and Lovaas

detail to the concern about subject assign-ment as well as other possible problemsassociated with the original study. There arecertain additional questions that may beraised by this follow-up investigation:

1. The experimental group was olderthan the control group at the time of thisfollow-up evaluation. We explained this find-ing earlier and noted that data analysesindicated that it was unlikely that this agedifference reflected a bias in subject assign-ments.

2. The follow-up assessments for 17 ofthe lower functioning subjects in this studywere conducted by staff members from ourProject, who could have biased the testresults. However, as noted previously, acheck revealed no evidence of such a bias.

3. The Clinical Rating Scale, based on aninterview with subjects who had been clas-sified as normal-functioning in the originalstudy I has no norms or data on reliability andvalidity. However, we regard the interviewsimply as an extra check on whether theexaminers detected residual signs of autismor other behavior problems that were some-how overlooked in the three other Cwell-standardized) measures in the study andtheir 30 subscales. We do not regard theinterview as an instrument that by itselfyields conclusive results. No other interviewthat suited our purposes currently exists. Infuture investigations, we plan to use aninterview that Michael Rutter and his associ-ates are now developing for the purpose ofdetecting of residual signs of autism in indi-viduals with average intelligence.

4. As in most long-term follow-up stud-ies, we had some missing data, However,there is no evidence that the niissing datawould have changed the overall results.

S. In our analysis of the best-outcomegroup, we noted that the group averagesdeviated from "normal" on one subscale ofthe Personality Inventory for Children andon the Clinical Rating Scale. We then attrib-uted this deviance to the extreme scores ofone subject rather than to general problemswithin this grO\lp. We recognize that group

370 Autism and Early Intervention

averages are seldom interpreted this way.However, as statisticians and methodolo-gists have pointed out (e.g. , Barlo,w & Hersen,1984), there are many times when groupaverages represent the performance of fewor no subjects within the group. This was oneof those times, as is clearly shown by the dataon individual subjects (Tables 2, 3, and 4).Deviance w~; found almost exclusively inone subject, not evenly distributed across allsubjects, and 'we have presented the resultsaccordingly.

The most important void for research tofill at this time is replication by independentinvestigators who employ sound method-ologies. Given the objective asse'ssment in-stroments that we used and the detaileddescription that we have provided of l:hetreatment (Lovaas et al., 1980), S\lch a repli-cation should be possible. Hmwever , thetreatment is complex and to replicate itproperly, an investigator probably needs topossess (a) a ~;trong foundation in learoingtheory research; (b) a detailed knowledge ofthe treatment manual we used; (I:) a super-vised practicum of at least 6 monl:hs in one-to-one work with clients who have develop-mental delays, emphasizing discriminationlearning and building complex langua/~e;and (d) a commitment to provide 'iO hours ofone-to-one treatment to client per week, SOweeks per year, for at least 2 years. Our best-outcome subjects an required a minimum of2 years of intensive treatment to achieveaverage levels of functioning (another indi-cation that those subjects had pervasivedisabilities and were not merely non-compliant).

A second void to fill concen1S the ma-jority of children who did not benefit to thepoint of achieving normal functioning withintensive treatment Perhaps an earlier startin treatment would have been aJI that wasneeded to obtain favorable outcomes withmany of these children. More pessimistically,perhaps such childl1en require ne'w and dif-ferent interventions that have "fet to bediscovered and implemented. In any case, itis essential to develop more approprialte

services for these children.Finally, a rather speculative but promis-

ing area for research is to detennine theextent to which early intervention altersneurological structures in young childrenwith autism. Autism is almost certainly theresult of deficits in such neurological struc-tures (Rutter & Schopler, 1987). However,laboratory studies on animals have shownthat alterations in neurological structure arequite possible as a result of changes in theenvironment in the first years of life (Sirevaag& Greenough, 1988), and there is reason tobelieve that alterations are also possible inyoung children. For example, children under3 years of age overproduce neurons, den-drites, axons, and synapses. Huttenlocher(1984) hypothesized that, with appropriatestimulation from the environment, this over-production might allow infants andpreschoolers to compensate for neurologicalanomalies much more completely than doolder children. Caution is needed in gener-alizing from these findings on average chil-dren to early intervention with children withautism, particularly because the exact natureof the neurological anomalies of childrenwith autism is unclear at present (e.g. , Rutter& Schopler I 1987). Nevertheless, the findingssuggest that intensive early intervention couldcompensate for neurological anomalies insuch children. Finding evidence for suchcompensation wQuld help explain why thetreatment in this study was effective. Moregenerally, it might contribute to an under-standing ofbrain-behavior relations in youngchildren.

.

~

Q

References

Barlow, D.H.,&Hersen,M.(1984). Single caseexperimental design: Strategies for studyingbehavior change (2nd ed.). New York: Per-gamon Press.

Bettellielm, B. (1967). The emptyforlress. NewYork: The Free Press.

DeMyer, M. K., Hlngtgen,J. N., &Jackson, R.K. (1981). Infantile autism reviewed: A de-cade of research. Schizophrenia Bulletin, 7,

388-451.Dunn, L. M. (1981). Peabody Pi(;tUI-e Vocabu-

lary Test-Revis,~d. Circle River, MN: AmericanGuidance Service.

Freeman, B.J., It.itvo, E. 1{., Needle(nan, R., .a:Yokota, A. (1'~8S). The stability olf cognitiveand linguistic Jparameters in autisml: A 5-yearstUdy. Journa~' 01 the American A,r;ademy ofChild Psychiatry. 24, 290-311.

Huttenl0<:her, I'. a. (1984). Synapse elimina-tion and plasticity in developing h\Jlman cere-bral cortex. At1oaericanjournal ofMentalDeft-ciency, 88, 4~H96,

KamJn, LJ. (19'74). Tbesciencea~rpoliticsofI,Q. New York: WjJey.

Kanner, L. (197:0. Follow-up study of II autis-tic children originally reported in 1943. jour-nal 01 Autism ,and Childhood Schi:~opbrenia,1, 119-145.

Kazdin. A. (19811»). Research design in clini(;alpsych,%gy, New YO£k: Harper & Row.

Kendalll, P. C., ~I: Norton-Ford, J. D. (1982).Therapy outcome research methods. In P. c.Kendall & J. N. Butcher (Eds.), H~ndbook ofresearch methj>ds in clini(;ai psychology (pp,429-460). New' York: Wiley.

Leiter, R. G. (19~i9). Part I of the manual for the1948 revision of the Leiter International Perfor-mana~ Scale: ]~vidence of the reli~Lbility andvalidity of the Leiter tests. Psychology ServiceCenterJournaj~ 11, 1-72.

Lord, c., .a: SChopler, E. (1989). The role of ageat assc~ssment, t:levelopmentallevel, and test inthe stability oj: intelligence scores in youngautistic chjJdreJ:l. Journal of Autism and Devel-opmental Disorders, 19, 483-499.

Lotter, V. (197:S). Follow-up studies. In M.Rutter & E. Sdllopler (Eds.) , Autis,n: A reap-pra~ai 01 concepts and treatment. London:Plenum Press.

Lovaas, 0. I. (1~'87). Behavioral treatment andnorm:u educatiional and intellectual function-ing in young autistic children. joun2a1 ofCon-suiting and cnni(;t:il Psychology, 5!.;, 3-9.

Lovaas, O.I.,Ac]~ennan,A. B.,Alex:lnder, D.,Flrestone, P., PerkblS, J., .a: Young, D.(1980). Teaclling developmental~v d~abledchildrr!n: The lne book. Austin, TX: Pro-Ed,

Lovaas, 0.1., K(Joegel, R. L, SlmIIlofas,J. Q., .a:Long, J. S. (V~73). Some generalization andfollow-up me;asures on autistic children inbehavior therapy. journal of Applied BehaviorAnalysis,6, 131-166.

Lovaas. 0. I., & Leaf, a. L. (1981). Five video

J McEachin, Smith, and Lovaas 371

I

r

Psychoeduc,ltional eva/uation oJt' schoo/-agedchildren wil'h /ow-incidence di~orders (pp.lSS-l80). Orlando, FL: Grune & Stranon.

Simeonnson, R.J., Olley,J. G., 4 Rosenthal,S. L (1987). Early intervention for childrenwith autism. In M.j. Guralnick & F. C. Bennett(Eds.), The ~tfectiveness of ear/y interoentionfor at-n:sk a:nd handicapped cl,i/dren (pp.275--296). Orlando, FL: Academk Press.

Slrevaag, A. M. , 4 Greenough, W. T. (1988). Amultivariate statistical summary of synapticplasticity measures in rats expo!;ed to com-plex, social arid individual environrnents. BrainResearch, 441, 386-392.

Sparrow. S. S., BaUa, D. A., a:. Clc(hettl. D. V.(1984). IntenliewEditionSuroeyFormManual.Circle Pines, lvlN: American Guidance Service.

Spin, H. H. (1~.86). The raising of;inte//igence.Hillsdale, Nj: Erlbaum.

Waterhouse, L., 4 Fein. D. (198~). Develop-mental trenw; in cognitive skills J'oc childrendiagnosed as :lutistic and schizophrenic. ChildDeve/opment, 55, 236-248.

WecMler, D. (1974). Manua/fortj!le WechslerIntelligence Scale for Chi/dren-REwised. NewYork: Psychological Corp.

Win, R. D., La.char, D., K.linediOJit, J. K.. 4Seat, P. D. (I~J77). Mu/tidimensional descrip-tions of chi/d persona/ity: A ma"lua/ for thePersonality 171~tory for ChildTe7l. Los Ange-les: Western Psychological Service~.

Zigler. E., a:. Trlckett, P. K. (1978). IQ, socialcompetence, and evaluation of E~arly child-hood intervention programs. Am,~rican Psy-ch%gist,33, 789-798.

tapes for teaching developmentally diS'abledchildren. Baltimore: University Park Press.

Lovaas, 0. I., a Smith, T. (1988). Intensivebehavioral treatment with young autistic cl1iI-dren. In B. B. Lahey & A. E. Kazdin reds.).Advances in clinical child psychology(Vol. 11,pp. 285-324). New York: Plenum Press.

Lovaas, 0. 1., Smith, T ., a McEachln, J. J.(1989). Clarifying comments on the youngautism study: Reply to Schopler, Short andMesibov. Journal of Consulting and ClinicalPsychology, 57, 165-167.

McEachin, J. J. (1987). Outcome of autiS'ticchildren receiving intensiw behavioral treat-ment: psychological status 3 to 12 years later.Unpublished doctoral dissertation, Universityof California, Los Angeles.

Newsom. C., a Rlncover, A. (1989). Autism. InE.J. Mash&R. A. Barkley(Eds.), Treatmentofchildhood disorrJers(pp. 286-346). New York:Guilford Press.

Butter, M. (1970). Autistic children: Infancy toadulthood. Seminars in Psychiatry, 2,435-450.

Butter, M. (1985). The treatment of autisticchildren. Journal of Child Psychology & Psy-chiatry, 26, 193-214.

Butter, M., a Schopler, E. (1987). Autism andpervasive developmental disorders: Conceptsand diagnostic issues. Journal of AutiS'm andDevelopmentalDisorrJers, 17, 159-186.

Schopler, E., Short, A., a Mesibov, G. (1989).Relation of behavioral treatment to "normalfunctioning": Comment on Lovaas. Journal ofConsulting and Clinical Psychology, 57,162-164.

Short, A., a Man:us, L. (1986). Psychoeduca-tional evaluation of autistic children and ado-lescents. In S. S. Strichart & P. Lazaros (Eds.),

Received: 5115191; rlrst decision: 10116191; acce~ed:1/23/92.

372 Autism and Early Intervention