Participants were recruited for the study beginning in May of 2018 and the final water kit was evaluated in October. The best response occurred over the summer when school was out. The results are summarized by Region in Table 3. To recap, the Southwest region included Central Arizona and Southern California, both served by a combination of Colorado River water and local sources. For this region, sulfate levels are much higher than the national norm since Colorado River water is typically in the 250 mg/L range. The Lowest and Highest Prevalence regions each consisted of the eight states with the lowest and highest rates of autism per Department of Education data. Finally, all of the regions were merged with participants from the rest of the country for a look at the USA as a whole.
Within the table, each region is sub-divided into Severity groups. The severity ratings are simply Mild, Moderate and Severe along with a row for All severity groups within a region combined. Severity was a subjective measure noted by each mother in the survey. It is not an official medical opinion although it was certainly influenced by the professionals consulted by mothers.
The results table shows n as the number of participants in each subgroup. The Tap Water column represents the average sulfate concentration of tap or well water reported in mg/L. Water Mix is the average sulfate concentration of the water mixture actually consumed by the mothers. It is a weighted average based on the number of glasses per day estimated for each type of water. It differs from the tap value because drinking water is commonly filtered and/or supplemented by bottled water. The standard deviation is listed next. The final columns switch to milligrams as units for the total sulfate reported from beverages plus water along with the standard deviation. Bev & Water was calculated by adding the sulfate concentration times the amount that was drunk for each beverage and type of water. All sulfate data points are means of n samples.
To better understand the results table, it helps to note a few facts about water and beverages. The 2003 EPA report estimates the median public water system across the country to have a sulfate concentration of 24 mg/L. The range is quite wide from zero to above 600 mg/L indicating that the mean would be higher than the median, although no estimates are given. The sulfate obtained from 2 L of median public water would be 48 mg. Beverages like coffee, tea, milk and juice are discussed by a US Department of Agriculture data brief titled Beverage Choices of U.S. Adults. Using this information (but reducing soda and alcohol for pregnant mothers) provides an estimate of 90 mg of sulfate per day from beverages (LaComb et al. 2011). Combining, 138 mg would be an estimate for sulfate from beverages plus water for a typical pregnant woman.
Results in Graphical Form
The results table is best visualized with the help of line graphs. Figure 1 plots Tap Water sulfate concentrations for each of the four study regions as a function of autism symptom severity. The data lines are widely separated but unsurprising. The Southwest shows relatively high sulfate simply because it is served by Colorado River water which is naturally high in minerals. The High Prevalence region reports the lowest sulfate values in this study, reaffirming water quality reports from major cities for these states. The Low Prevalence region indicates high sulfate for mild conditions, dropping sharply for more severe autism. While the USA as a whole shows a flat, mid-range plot with little apparent association with symptom severity.
Figure 2 plots the sulfate concentrations for the Water Mix actually consumed by the pregnant mothers. This includes tap or well water, filtered and unfiltered, along with a variety of bottled waters. Again, each of the four regions is plotted against autism severity. These curves are surprisingly different from tap water alone. Three of the four regions suggest a trend, hinting that mild symptoms and higher sulfate may be related. High Prevalence is the outlier, more or less flat but on the low end of sulfate. Standard deviations for the data points are shown in Table 3.
The sum of sulfate from Bev & Water is presented in Fig. 3. These are the beverages, tap water and bottled waters that were drunk by the participating mothers. In this case, sulfate is not a concentration but the mg weight of the sulfate consumed daily. It was calculated from the number of 8 oz glasses reported by each mother for various types of beverages and water. As in previous figures, each of the regions is plotted against autism severity. All of the regions now show a monotonic, decreasing function clearly associating severity with lower levels of sulfate. These relationships are examined in the regression analysis that follows.
Regression, Correlation and Comparison
The appearance of the plotted data suggests a reasonable correlation between sulfate consumed by pregnant mothers and the resulting severity of autism shown in their children. Of course, this assumption needs to be tested for statistical significance. Table 4 presents the relevant data for the Beverage plus Water graph. Each of the four regions is characterized by n, the number of data sets, and by the Mean ± SD (standard deviation) of sulfate for all levels of autism severity. Then a linear regression is performed on the data sets to minimize the sum of the squared errors. To perform the numerical calculations, autism severity is mapped in the following manner: mild = 1, moderate = 2 and severe = 3. The resulting Pearson Linear Correlation Coefficient is noted as r in the table along with a description of the Strength of Correlation. In the Strength column, p is the probability that the correlation is a result of statistical chance and therefore invalid. For small sample sizes in our study, p is not insignificant but drops to < 1% for the full USA.
The correlations are negative indicating sulfate is inversely related to autism severity, lower sulfate resulting in higher severity. However, the relationship is not strong, as the correlations are typically rated as low. Such results might be expected noting the wide spread of data. But even a low signal seems significant given the unknowns in the study. The genetics of the mothers and children were completely unknown, other than the generalization that autism resulted. Sulfate from food sources was not tracked, only sulfate in drinking water and beverages. Since the typical age of the child in the study was 6 years, memory errors may have occurred. And the water tested was therefore many years out of date. Finally, the mothers reported typical water and beverage consumption that may have been representative, but this may have varied over the course of a full pregnancy. With all of this uncertainty, the actual survey results are surprisingly clear. It is worth noting that the study was conducted in four dissimilar regions with widely differing levels of sulfate in tap water and very different rates of autism. In essence, this was four studies rolled into one. And all four showed a similar correlation between sulfate and autism.
A graph is shown in Fig. 4 to better visualize the relationship, data spread, regression and correlation. It is the plot of sulfate in Beverage plus Water vs Autism Severity for data covering the full United States. Severity mapping to the sequence 1, 2, 3 is depicted along with single standard deviation error bands and the resulting linear regression line. For this set of 86 samples, the correlation is r = − 0.32 with the probability of a statistical flaw being p < 0.01, < 1%. In a single figure, this is a snapshot of the results of our study.
Welch unequal variance t-tests can be applied to tap water, testing the statistical validity of sulfate differences between regions. To focus on water alone, we include all reported values of sulfate in tap water without regard to autism severity. The Southwest region was selected for its link to Colorado River water and probable high mineralization. Our survey reports a Southwest sulfate mean of 151 mg/L compared to 59 mg/L for the other regions combined. The probability of this difference as untrue is p = 0.00037, confirming our assumption of higher than average sulfate levels for the Southwest. The Southwest will be examined further in the discussion section.
The eight states with the lowest rates of autism form the Low Prevalence region which reports a tap water sulfate mean of 94 mg/L. The eight states with the highest rates of autism form the High Prevalence region with a sulfate mean of 26 mg/L, just 28% of the Low Prevalence average. Applying a Welch t test, the null probability of this difference as untrue is p = 0.059. Though on the cusp of the usual 5% benchmark, this is an indicator of the importance of sulfate considering tap water values only represent the water that is available, not the water mix actually consumed. It is significant that the prevalence regions were selected using Department of Education autism data alone. Yet they show a definite association with local tap water sulfate concentrations.