Introduction

The measles, mumps, and rubella (MMR) vaccine has been suggested as a possible cause of a “new variant of regressive autism” (Wakefield et al., 1998). In Japan, the MMR vaccination was introduced in April 1989, when the Japanese government recommended that the MMR vaccine or a monovalent measles vaccination be given once to toddlers between the age of 12 and 36 months. In Japan, only one shot of MMR was included in the immunization schedule. The monovalent mumps and rubella vaccine remained the optimal choice of vaccine for those who did not participate in the MMR program. In Japan, the MMR vaccine contained AIK-C (measles), Urabe AM9 (mumps), and To-336 (rubella) strains. However, soon after the introduction of the MMR program, there were several cases of aseptic meningitis, which may have been caused by the Urabe strain of the mumps virus (Sugiura & Yamada, 1991). As a result, in April 1993, the Japanese government ceased extensive inoculation with MMR (Takahashi, Arai, Tanaka-Taya, & Okabe, 2001; Takahashi et al., 2003).

Wakefield et al. (1998) have postulated that the MMR vaccination causes a variant of autism associated with developmental regression and bowel symptoms. This hypothesis predicts that the incidence of regressive autism should be higher in children given the MMR vaccine, and that, in the case of Japan, the incidence of regression in autism should have increased following the introduction of the MMR program and decreased after it ceased (Taylor et al., 2002). In order to investigate whether the MMR vaccination is associated with “regressive autism”, we examined the rate of autism spectrum disorders (ASD)Footnote 1 involving regressive symptoms in children who did and those who did not receive MMR during the MMR period (i.e. a case–control comparison), and the change over time in the proportion of children who showed regressive symptoms across the pre-MMR, MMR, and post-MMR periods in Japan (i.e. a time-series design). The case–control comparison has the advantage of using data at the individual level but the disadvantage of uncertainty regarding the family decision on whether the child should be vaccinated. The time series design has the major advantage of avoiding the possible confound of that uncertainty (because the decision to withdraw MMR was taken by the Government and not the family). However, it has the disadvantage of needing to use aggregate data. The combination of these two approaches with their different patterns of strengths and limitations provides a double test of the MMR hypothesis.

Methods

Subjects

The study analyzed data from clients of the Yokohama Psycho-Developmental Clinic (YPDC), located in Yokohama, Japan. The YPDC, a private child psychiatric clinic specializing in developmental disorder, opened in April 1997. The YPDC accepts only patients with suspected developmental disorder. Most patients reside in the area of Yokohama and Tokyo, but some are referred from elsewhere in Japan. The YPDC has a close relation with the many parental organizations advocating for autism in Japan and the YPDC has become recognized as a centre for ASD. For this reason, the proportion of clients with ASD is very high.

Diagnosis

Prior to February 2000, intake of clients at the YPDC involved an initial assessment process. In the first instance, one of several child psychiatrists on the team conducted assessments using the Diagnostic and Statistical Manual (DSM-IV) (American Pyschiatric Association, 1994). Thereafter in most cases, a clinical psychologist conducted an intelligence test. At intake, there was no discussion on diagnosis among psychiatrists on the team. After intake, psychiatrists followed most of the patients once or twice a month. The majority of clients on the caseload were seen by the first author and some other clients were seen by other doctors. There were no reliability data on diagnoses made by psychiatrists, but as all the doctors on the team have worked with the first author either at the same child psychiatric hospital or at YPDC, each member has received training using a common concept of diagnosis.

In 2000, the YPDC became an independent private clinic. Since then, initial assessment procedures have changed. A team of one child psychiatrist and several clinical psychologists now conduct full assessments in the course of one day. The psychiatrist interviews the parents for several hours. Clinical psychologists conduct intellectual or developmental tests, such as the Japanese version of Psycho-Educational Profile-Revised (Shopler, Reichler, Bashford, Lansing, & Marcus, 1990), the Japanese version of Wechsler Intelligence Scale for Children—Third Edition (Azuma et al., 1998), and the Tanaka-Binet test (a Japanese modification of the Stanford–Binet test) (Tanaka, 1987).

Diagnosis is based on clinical observation, interviews with the client, and a parent interview regarding the developmental history of their child’s social, behavioural, and communicative functioning. Three child psychiatrists, each with 10 or more years of experience in child psychiatry, independently make a diagnosis of ASDs in the current system. Although there are no reliability data about the diagnoses made by the three child psychiatrists, they discuss the diagnoses and treatment plans as often as possible, and they share ideas about clinical principles. Therefore, the psychiatrists at YPDC essentially apply similar criteria to their diagnoses, and draw from shared clinical principles. All the charts reviewed in the present study reflect diagnostic processes from either the pre- or post-February, 2000 period described above.

Source of Data Concerning Regression

Before this research began, in all cases, historical data from a standard questionnaire regarding the patient’s developmental, behavioural and medical history that had been completed by parents was archived in a database. The data concerning regression were extracted from this source. The standard questionnaire we used as a clinical tool probed parents’ perceptions of their child’s development, behaviour, medical history, and present status. We did not examine the reliability of this instrument because, prior to this investigation, we had been routinely using the same questionnaire for clinical purposes only (Table 1).

Table 1 Items on the questionnaire concerning regression

In this study, the question of whether or not regression occurred was the dependent variable. The definition of regression used in this study is the same as that proposed by Taylor et al. (2002), who defined ‘regression’ as “a documented deterioration in any aspect of development or reported loss of skills, however transient”. In this study it was determined that regression had occurred when language or any other skills were reported, on the questionnaire completed by parents, to have regressed. The questionnaire was designed such that we were able to study regression in language skills more systematically than in other areas, though we noted regression of skills other than language whenever possible.

Source of Data Concerning the MMR Vaccination

Information as to whether the child had been vaccinated with MMR was based on a copy of the Maternal and Child Health (MCH) handbook, which was routinely attached to each patient’s file. The MCH handbook, which is given to all mothers by the government, is a highly reliable record of early development, health, and immunization. Health professionals (e.g., public health nurses and paediatricians) record most of this information. These MMR vaccination records were entered into a computerized database for analysis.

Statistical Analysis (see Fig. 1)

Data were analyzed from all patients who had been clinically assessed at the YPDC between April 1997 and December 2002. Included in the study were all patients with the clinical diagnosis of ASD. When we began our investigation, a total of 1,053 children and adults had the diagnosis of ASD. We selected the patients who were born between 1976 and 1999 for the study. There were 907 patients with ASD identified in this period. Since this study is on MMR and regression in Japan, we excluded three children who received the MMR vaccine in foreign countries. Thus, 904 cases remained. After reviewing the questionnaires, out of those 904 cases, regression was reported in 325 cases. Out of 325 cases of regression, 230 cases indicated language regression (72 cases of these had also regression in skills other than language), 167 cases indicated regression in skills other than language (again, 72 cases had also regression in language), and 72 cases indicated regression in both language and other skills.

Fig. 1
figure 1

The area of regression in 325 cases among 904 patients with ASD

The MMR vaccine program was implemented from April 1989 to the end of April 1993.Therefore, children in the last generation that received the MMR vaccination (the youngest children now) were one year old at the end of April of 1993, i.e. these were children who were born in April of 1992. On the other hand the older children in the MMR inoculation period were under 4 years old at the beginning of the MMR inoculation period, i.e., these children were born in April, 1985. We labelled the cohort that could have received an MMR vaccine the “MMR Generation”. Therefore, the MMR Generation should have been born between May 1985 and May 1992. However, our preliminary analyses showed that all the children in the MMR Generation who received the MMR vaccine were born between January 1985 and December 1991. Thus, we defined the MMR Generation as those born between January 1985 and December 1991.

We classified the patients into three groups according to the chance of having received an MMR inoculation. The Pre-MMR Generation (n = 113) were those patients born between January 1976 and December 1984 who had not received the MMR vaccine. The MMR Generation (n = 292) were born between January 1985 and December 1991 and had a chance of receiving the MMR vaccine. The post-MMR Generation (n = 499) were within the age of 1–3 years old after 1993, when the MMR immunization program was terminated (see Table 2, Fig. 2).

Table 2 Date of birth and number of subjects with ASD
Fig. 2
figure 2

The numbers of cases with the information of regression and MMR status in the MMR-program-related three generations

Out of our sample of 904 cases, there were 46 patients whose information on regression was unknown because parents had provided no information on questionnaire items that related to regression. That is, of these 46 cases, 13 were in the pre-MMR Generation, 17 in the MMR-Generation, and 16 in the post-MMR Generation, respectively. We excluded these 46 patients from the analysis as there was no way to verify whether regression had or had not occurred. Thus, the number of cases to be analyzed in each group was reduced as follows: there were 100 in the pre-MMR Generation whose parents provided information on regression, 275 in the MMR Generation, and 483 in the post-MMR Generation.

We further classified the MMR Generation with the information on regression (n = 275) into three subgroups: (1) a group that had apparently received MMR (n = 54); (2) a group that had not received MMR (n = 132) and (3) a group whose vaccination history was unclear because the MCH handbook was lost or because the recorded description of vaccination was unclear or hard to read (n = 89). Thus, we excluded these patients from the analysis. In total, then, there were 54 cases of children who had received an MMR vaccine, and 132 who had not.

Analysis 1

Within the MMR Generation, we analysed the odds ratio for the association between regression and MMR exposure using the MMR-subgroup of children who had received the vaccine as the reference. We excluded the data of the MMR-vaccination-status-unclear group.

Analysis 2

Using all data across the three Generations, we computed an odds ratio for the association between regression in ASD and exposure to MMR by combining data from three periods. All odds ratios (ORs) are presented with 95% confidence intervals (CIs) and categorical variables were compared by the two-sided Fisher’s exact test.

Analyses 3,4 and 5

In order to have time trends data, three groups (i.e. pre-MMR, MMR, post-MMR Generation groups) with and without regression were analysed. We computed the odds ratio for the association between ASD and regression using the MMR Generation as the reference group. That is, comparison of the pre-MMR and MMR Generations (i.e., analysis 3), comparison of the MMR Generation and post-MMR Generations (i.e., analysis 4), and comparison of the MMR Generation and the pre-MMR and post-MMR Generations combined group (i.e., analysis 5). All these analyses were carried out to investigate the hypothesis that MMR has been responsible for the claimed rise of “regressive autism”.

Results

Analysis 1

Compared with the MMR-subgroup that did not receive the vaccine during the MMR era, the OR for those in the MMR group who received the vaccine was 0.744 (95% CI: 0.349–1.517, P = 0.490, Fisher’s exact test). That is, although the rate of regression was lower in those receiving MMR during the MMR era (27.8% vs. 34.1%), this difference was not statistically significant. (Table 3)

Table 3 Proportion of regression among those receiving and those not receiving MMR in-MMR period and odds ratio

Analysis 2

Compared with the children who had not received the MMR vaccine, the OR for the children who had received the MMR in three Generations was 0.626 (95% CI: 0.323–1.200, P = 0.146, Fisher’s exact test). That is, the lower rate of regression in the children receiving MMR (27.8% vs. 38.0%) fell short of statistical significance (Table 4).

Table 4 Odds ratio between MMR done and MMR not done patients with ASD

Analyses 3,4 and 5

Compared with the pre-MMR Generation, the OR for the MMR Generation was 1.075 (95% CI: 0646–1.791), the OR for the post-MMR Generation was 0.832 (95% CI: 0.605–1.144), and the OR for the pre-MMR and post-MMR Generations combined was 0.868 (95% CI: 0.638–1.182). That is, the rate of regression did not vary across the pre-MMR, the MMR, and the post-MMR eras (Tables 5, 6, 7).

Table 5 Number of cases and rate of regression in MMR and pre-MMR Generations and ORs (95% CI) excluding “regression unknown”
Table 6 Number of cases and rate of regression in MMR and post-MMR Generations and ORs (95% CI) excluding “regression unknown”
Table 7 Number of cases and rate of regression in MMR and pre- + post-MMR Generations and ORs (95% CI) excluding “regression unknown”

Discussion

This study focused strictly on the hypothesis that the MMR immunization is distinctive in leading to a new variant form of autism with developmental regression, and that, as a result of the MMR vaccination, there has been a recent increase in the rate of “regressive autism”. Within the MMR era, the rate of regression in those who received MMR was not higher than those who did not. Moreover there was no indication that the rate of regression in ASD was higher during the era when MMR was used, compared to the “before” period, and “after” period, and the “before and after” periods combined.

To test the alleged link between the MMR vaccination and regressive autism, Japan has several advantages. First, in Japan, the MMR vaccination program was conducted only for children born between 1985 and 1991. Thus, we can examine the rate of regression in ASD before, during, and after the period of the MMR vaccinations. Our MMR Generation group contains many children who were unexposed to MMR, and thus we must be cautious in comparing the three Generations. However, if the MMR vaccination is related to “regressive autism”, regression in the development of children with autism should be more common in the MMR Generation than in the pre- and post-MMR Generations. Our results showed no such relation. The results of analyses 1 and 2 taken together disconfirm the hypothesis that the MMR vaccination causes regression in ASD.

The type of cross-generational analyses presented in this paper could only have been conducted in Japan, given the MMR inoculation history described in this paper. Unlike European countries and the U.S.A., few Japanese parents are aware of the alleged link between the MMR vaccination and autism/ASD, for there has been no newspaper publicity on the issue. Furthermore, Japan had already terminated the MMR immunization at the time that Wakefeld et al. (1998) suggested a relationship between regressive autism and the MMR immunization. As a standard part of our data gathering, parents answered questions in relation to regressive symptoms. Thus, both clinicians and parents were relatively free from bias concerning any link between autism and regression. In other words, as parents who filled out our questionnaires were blind to the possible association of MMR and regression, the data from these Japanese parental reports hold more credibility than any such parental reports from the U.S.A. or the U.K. might hold. It is probable that, in Europe and in the U.S.A., publicity over MMR has made people more aware of regression, and more likely to respond positively to questions about it.

Limitations

There are several limitations in this study. First, this study was conducted in a specialty private clinic. Thus it was not a population-based study, which raises the possibility of potential selection bias. Second, the sample size of those subjects who had actually received the MMR was relatively small. Third, there are several issues concerning definition and measurement of regression. The information on regression was totally dependent on parental report using an instrument, which asks limited questions about regression. The concept of regression we employed in this study is probably wider than “real” regression. Further, our study may have identified some “false positive” cases of regression, partly because questions were limited, and partly because we did not assess regression clinically. Through the three generations, the overall rate of regression in ASD was 37.9% (see Table 7.). Moreover, the rate of regression in autism or ASD varies, depending on the study. The reason for different prevalence rates is partly due to the diagnostic category of the sample (i.e., autism or ASD), or the definition of regression (speech loss, or the loss of any adaptive skill). The prevalence of regression in some previous studies has been as high as 49% in autism (e.g., Hoshino et al., 1987), 30% in ASD (Tuchman & Rapin, 1997), and 26% of speech loss in ASD (Kurita, 1996). More recently, Taylor et al. (2002) reported that the rate of regression in autism and atypical autism was 25%, and Fombonne and Chakrabarti (2001) reported that 15.6% of children with ASD had experienced regression. As compared with recent surveys, our finding that regression occurs in 37.9% of patients with ASD is rather high. One of the reasons for the rather high figure in this study may be that we judge the existence of regression only by parental report and that the definition of regression was rather lax. However, Fombonne and Chakrabarti (2001) defined regression strictly, they employed a similar method to our study, and they also disconfirmed the hypothesis that the rate of regressive autism increased after the MMR program started in the U.K.

Regarding the diagnosis, we used the ASD label and did not differentiate by subcategory, as it was difficult to elicit information before the child was 3 years old. Further, it would have been even more difficult to conclude that the developmental level before 3 years of age was within in normal limits or not. This information is crucial to making a differential diagnosis of Autistic disorder, Asperger disorder or pervasive developmental disorder not otherwise specified using the DSM-IV system. Therefore we analysed the link between ASD and regression rather than the possible link between autism and regression.

Several epidemiological studies designed to investigate the link between MMR and autism have found no such relationship. All these studies so far were conducted in the U.K. (Fombonne & Chakrabarti, 2001; Kaye, del Mar Melero-Montes, & Jick, 2001; Taylor et al., 1999, 2002), and in Europe (Madsen et al., 2002; Makela, Nuorti, & Peltola, 2002). Our results cast additional doubt on the hypothesis that developmental regression in children with ASD is associated with MMR.