Autism and Vaccines: Thesis Statement

Abstract

Almost two decades ago, a group of scholars led by a British doctor claimed in The Lancet that the MMR vaccine caused autism in 8 children. Although a substantial body of epidemiological evidence on the safety of the MMR vaccine has accumulated since then, measles outbreaks continue to occur in the U.S. and at least some of those outbreaks were attributed by the media to the anti-vax movement spreading misleading information about vaccines. This research proposal suggests using a novel empirical approach and a particularly rich dataset to re-examine the causal relationship between the MMR vaccine and autism. In particular, stratifying the sample based on the propensity score and controlling for not only children’s but also their parents’ health and demographics will make it possible to address potential sources of self-selection bias that might otherwise prevent a causal interpretation of the relationship between completion of vaccinations and autism in the U.S. population.

Introduction

Public concern that the measles, mumps, and rubella (MMR) vaccine can cause autism can be traced to a 1998 study that was published in The Lancet by the British doctor Andrew Wakefield and his 12 co-authors (Wessel, 2017; Godlee et al., 2011). The study sample was only 12 children, there was no control group, and the paper itself was retracted in 2010 over Wakefield’s failure to disclose a conflict of interest (Wessel, 2017; Plotkin et al., 2009). Since then, multiple studies conducted in various countries have found no evidence of an association between the MMR vaccine and autism. More recently, the National Academy of Medicine (formerly known as the Institute of Medicine) has further determined that the MMR vaccine prevents the rubella disease, which was linked with autism (Dudley et al., 2018).

Nevertheless, almost a decade after the controversial study was redacted, the popular myth that vaccinations might cause autism persists (Bennett et al., 2018). In 2019, a Wall Street Journal article raised concerns about thousands of schools not meeting the recommended 95 percent MMR vaccination rate (Abbott et al., 2019). Although the exact reasons for low immunization rates in some communities are unknown, in other communities public health officials and the media linked low immunization rates to the anti-vaccine culture. For example, when discussing low vaccination rates in Oregon, a public health official noted an increase in exemptions to immunizations from 1 to 7.6 percent between 2000-2017 and attributed that increase to the anti-vaccine movement (Leventhal, 2018). Similarly, a 2018 Washington Post article blamed the 2017 measles outbreak in Minnesota on anti-vaccine activists, alleging that they spread misinformation about the vaccine in an immigrant community, which led to 75 measles cases among the children of Somali immigrants (Sun, 2018).

In view of the persistence of anti-vaccine myths, it is important to continue to re-examine the MMR vaccine-autism link with novel methods and/or data. This study briefly reviews the literature on the association (or its lack thereof) between the MMR vaccine and autism and proposes a novel empirical strategy to re-examine that relationship. In particular, it proposes applying propensity score matching methods, which have been used in a variety of contexts including medical studies (Austin, 2008), to investigate whether autism is caused by the MMR vaccine in the U.S. population. This approach will make it possible to minimize, if not eliminate, self-selection bias that might make children who receive vaccinations and children who do not different along the unobservable dimensions. The data will come from the National Health Interview Survey, which collects detailed health information using a nationally representative sample of U.S. children and adults annually and includes questions about autism diagnosis and immunization history (CDC, n.d.). The use of this rich health data will make it possible to take into account not only children’s health status and socio-demographic characteristics but also their parent’s health status and socio-demographic profile. Since certain parental characteristics and behaviors are believed to be associated with autism, such as increased parental age or the mother’s exposure to certain drugs or chemicals during pregnancy (WebMD, 2019), and these parental characteristics might be associated with parents’ decision to vaccinate their children, controlling for parental variables is essential to mitigating potential self-selection bias.

Literature Review

Evidence against Wakefield’s findings on the link between the MMR vaccine and autism started piling up as soon as that article appeared in The Lancet. In 1999, Taylor and his co-authors conducted an epidemiological study and found that the introduction of the MMR vaccine in the UK in 1988 did not cause an increase in the prevalence of autism diagnoses. The authors concluded that even if an association exists, it must be so rare that they could not identify it in a very large regional sample. In 2004, the (then) Institute of Medicine examined more than 200 studies on the link between vaccines and autism, concluded that there is no relationship between the two, and recommended that funding for autism research be directed to other, more promising areas than the non-existent association between autism and the MMR vaccine (Institute of Medicine, 2004; Gross, 2009). However, the debate had not been settled.

Just two years after the publication of the Institute of Medicine’s comprehensive review of the literature, Doja and Roberts (2006) included the latest evidence in their literature review and arrived at the same conclusion, namely, that no causal relationship between the MMR vaccine and autism had been established. Three years later, Miller and Reynolds (2009) evaluated the empirical evidence and also concluded that no credible causal association between vaccines and autism has been found. In 2019, Hviid et al. published one of the largest-scale research projects on the relationship between the MMR vaccine and autism. Using Danish population registries for their nationwide study, the authors confirmed the absence of a causal relationship between the MMR vaccine and autism.

Save your time!
We can take care of your essay

• Proper editing and formatting
• Free revision, title page, and bibliography
• Flexible prices and money-back guarantee

Place Order

Over the last decade, it has become clear that the scientific evidence on the harmlessness of the MMR vaccine has reached some critical mass and the literature started to shift from the vaccine-autism link to the study of parents’ perceptions of vaccine safety, including parental characteristics associated with parents’ hesitation and refusal to let their children receive vaccinations. For example, Freed et al. (2010) conducted a survey to elicit parental views on vaccines and found that although Hispanics were the racial group most likely to think that vaccines trigger autism, they were also the group least likely to refuse a vaccine recommended to their children by a doctor.

The novelty of the proposed study is twofold. First, it will use an economic rather than purely epidemiological approach that explicitly recognizes the self-selection issue arising when one attempts to study the MMR vaccine-autism link in the U.S. population. Second, not only will a wide range of variables related to children’s demographics and health be held constant in this study, but parental demographics and health status will also be controlled for. This thorough approach to ensuring that children in the treatment and control groups are similar along the observable and unobservable dimensions, including in terms of their socio-economic background, will minimize the role of any environmental factors that might be associated with both the onset of autism and the choice of parents to let their children be given the MMR vaccine, essentially eliminating self-selection bias.

Methods

This study will test the hypothesis that receiving the MMR vaccine raises the odds of receiving an autism diagnosis. Thus, the dependent variable is binary and equals 1 if the child has an autism diagnosis and 0 if the child does not have an autism diagnosis. The treatment variable is also binary, taking the value of 1 if the child has received the MMR vaccine and 0 otherwise.

The main methodological challenge in identifying a causal relationship between vaccinations and autism diagnosis in the U.S. population might be potential self-selection bias. Since the majority of states grant exemptions to vaccinations for religious reasons, often without requiring documentation (College of Physicians of Philadelphia, n.d.), and multiple states grant exemptions for philosophical beliefs (National Conference of State Legislatures, 2019), families can “self-select” into vaccinating their children. Furthermore, in states that do not grant exemptions to vaccinations, families can choose homeschooling for their children to avoid vaccinating them if they strongly object to vaccinations. If autism is triggered at least in part by environmental factors in early childhood or during pregnancy, and those factors are correlated with the family's decision to vaccinate their children, then comparing children who received vaccinations and children who did not might produce an incorrect estimate of the effect of vaccinations on autism. For this reason, it is important to use an empirical approach that can eliminate or at least mitigate the selection bias. Many of the existing epidemiological studies mentioned above try to make the treatment and control group similar by matching observations on key demographic variables such as age, race, and sex. This study, in contrast, will try to make the treatment and control groups (children who were given the MMR vaccine and children who were not) similar along many more dimensions.

The propensity score methods generally have two steps. The first step involves estimating a logit or probit regression in which the dependent variable is the treatment variable (Dehejia & Wahba, 2002). This logit or probit model should control for all observable and unobservable characteristics of the sample that might be associated with selection into treatment (Dehejia & Wahba, 2002). The second step involves using one of the many available propensity score matching algorithms to match each treated observation with one or multiple control observations and then using a regression or simply stratifying the sample on the propensity score and calculating the difference in means between the treatment and control units to estimate the effect of the treatment variable on the variable of interest (Austin, 2008).

The use of propensity score matching necessitates choosing a matching algorithm. Caliendo and Kopeinig (2008) assess the advantages and disadvantages of the four most commonly used matching algorithms: the nearest neighbor approach, Caliper and Radius approach, the kernel and local linear approach, and stratification. The authors conclude that each approach involves a trade-off between bias and efficiency, and recommend trying multiple approaches to test the robustness of findings to the choice of the matching algorithm. Baser (2006) compared the performance of seven matching algorithms in a medical study and concluded that the choice of the algorithm has a substantial effect on results. This study will use the stratification matching algorithm (the algorithm used by Dehejia & Wahba (2002) in the highly influential paper) and one alternative matching algorithm that will be chosen later to test the sensitivity of the results.