A Geostatistical Analysis of Possible Spirochetal Involvement in Multiple Sclerosis and Other Related Diseases

© Megan M. Blewett 2006

Megan.Blewett@att.net

Abstract

 

Zoonotic diseases, especially those with insect or arthropod vectors, are recognized public health problems.  This class of diseases includes West Nile Virus, Human Granulocytic Ehrlichiosis (HGE), Babesiosis, Rocky Mountain Spotted Fever, and Lyme Disease.  This study examines whether Multiple Sclerosis (MS), which is the most common primary neurological disorder of young adults, also belongs in this category.  Visual and geostatistical analyses of MS and Lyme reveal striking similarities between the two diseases.  Maps displaying each disorder’s geographic distribution by county reveal this overlap visually.  In addition, the statistical correlation between MS and Lyme deaths (specifically all arthropod-borne disease deaths) is significant at the state-level and highly significant at the county-level.  MS incidence is known to vary with latitude; the study’s statistical analysis reveals that Lyme Disease follows the same trend.  Discussion of possible biological explanations of these geographical and statistical trends is included in this article.  Significant correlations also exist with other diseases: on the state level, the correlation between MS and breast cancer is 0.330, and between MS and ALS (Motor Neuron Disease used in this study), the value is 0.618.  The control, external accident/injury, did not yield significant correlations.  Producing the maps and data required contacting all of the state epidemiologists in the nation for Lyme incidence data.  Compiling the data has resulted in one of the most comprehensive Lyme databases available to researchers.  The results of the visual, geostatistical, and biochemical analyses suggest common spirochetal involvement in MS and related diseases.

A Geostatistical Analysis of Possible Spirochetal Involvement in Multiple Sclerosis and Other Related Diseases

 

Introduction

 

Zoonotic diseases, especially those with insect or arthropod vectors, are well-recognized public health concerns.  Such diseases include West Nile Virus, Human Granulocytic Ehrlichiosis (HGE), Babesiosis, Rocky Mountain Spotted Fever, and Lyme Disease.  Multiple Sclerosis (MS) is the “most common primary neurological disorder of young adults” (Warren, 2001, page 1).  The National Multiple Sclerosis Society estimates that 400,000 people in the United States have MS (National Multiple Sclerosis Society, 2005).  The National Institute for Neurological Disorders and Stroke (NINDS) reports that the cause of MS is “linked to an unknown environmental trigger, perhaps a virus (NINDS, 2006a).  Although a viral cause of MS is the prevailing view, some researchers believe MS is a zoonotic disease caused by a spirochete and spread by an arthropod vector.  This study examines the spirochete hypothesis.

Spirochetal involvement in MS was a hypothesis gaining ground in Europe in the 1930s (Murray, 2005).  Unfortunately, most of the research in support of this hypothesis, as well as the researchers themselves, was lost during World War II.  A surviving researcher, Gabriel Steiner, published work after World War II that identified a spirochete, Spirochaeta Myelophthora, as the causal agent of MS with an unknown vector (Steiner, 1952; Steiner, 1954).  Some of those who worked with Steiner in the United States as well as other researchers hypothesize that MS and Lyme might be either: 1) the same disease; or 2) different diseases caused by two different spirochetes carried by the same arthropod vector (Mattman, 2001; Rubel, 2003; Fritzsche, 2005).

 

 

Figure 1.  Normalized Count of MS Deaths by County (1998 Deaths Divided by 1990 Census Population)

 

 

 

Figure 2.  Normalized Count of Other Specified Arthropod-Borne Diseases (OSABD) Deaths by County (1998 Deaths Divided by 1990 Census Population)

Geostatistical and biochemical analyses reveal many similarities between MS and Lyme.  Each is influenced by geography, and MS and Lyme overlap in this geographic distribution.  The author began to examine the relationship between MS and Lyme after being struck by the similarity of the distribution apparent in generated distribution maps of both diseases.  See Figure 1 and Figure 2.  There are also biochemical similarities.  NINDS (2006a) defines MS as “An unpredictable disease of the central nervous system … in which the body, through its immune system, launches a defensive attack against its own tissues … the nerve-insulating myelin.”  NINDS (2006b) also recognizes the neurological complications of Lyme, which usually occur in the second stage, and include “numbness, pain, weakness, Bell's palsy … visual disturbances, and meningitis symptoms … decreased concentration, irritability, memory and sleep disorders, and nerve damage in the arms and legs.”

Each of the disorders is characterized by damage to the blood-brain barrier (BBB) endothelium and subsequent increased barrier permeability (Pardridge, 1998).  Degradation of the barrier in Lyme patients involves bacterial breakdown of the collagen in the BBB basement membrane.  The method of degradation in MS is not known (Russell, 1997), though thickness of the collagen layer could be a factor for prevalence among certain ethnic groups.  For example, African-Americans have high levels of collagen and low rates of MS.  Both diseases also involve demyelination triggered by what can resemble an autoimmune attack against the myelin sheath.  Among MS patients, the mysterious increase in lymphocyte movement across the BBB could be in response to a bacterial invader.  Lastly, MS and Lyme disease share an inflammatory response, most likely the work of proinflammatory chemokines and cytokines(Rothwell, 2002).  The epidemiological and biochemical similarities suggest, but do not confirm a common bacterial basis for MS and Lyme.

The possibility of a common bacterial basis for both MS and Lyme is examined in this study using geostatistical analysis.  Such analysis combines descriptive and inferential statistical techniques with data visualization (cartographics).  The results have proven useful in understanding the etiology of many diseases including cholera, plague, malaria, smallpox, AIDS, and Lyme (Ormsby, 2001, Cliff, 2004; Koch, 2005;).  The hypothesis to be tested is that MS and Lyme Disease are triggered or influenced by a similar zoonotic spirochetal agent and spread by a tick-like vector.  If a common etiology exists, then a geostatistical relationship between Lyme and MS should be observed at either the state-level or the county-level or both.  The analysis can be improved by using a control variable (disease) and at least one other condition in which the causal agent or geographic distribution might be similar to that of MS.

The control variable in this study is accident/injury because this condition should be unrelated to a bacterial distribution.  The two diseases with a suggested bacterial cause or geographic similarity to MS are Breast Cancer (Cantwell, 1998) and Amyotrophic Lateral Sclerosis (ALS, Lou Gehrig’s Disease) (Agency for Toxic Substances and Disease Registry, 2003).

Methods

Comparing disease distributions requires a database of the incidence of the diseases under examination and their associated environmental variables.  The data collection process began with a search for an authoritative source of incidence and prevalence data for Lyme, MS, Breast Cancer, ALS, and accidents/injuries.  Deaths recorded with the Centers for Disease Control and Prevention (CDC) and other government agencies provide an incidence measure of the given diseases.  A useful dataset was found on TheDataWeb, which is an online set of data libraries.  The dataset, “Mortality – Underlying Cause-of-Death – 1998” (United States Bureau of the Census (Census Bureau), 2005b; CDC, 2005c), was accessed via DataFerret, a data mining tool (Census Bureau, 2005a; CDC, 2005a).  The United States Bureau of the Census (Census Bureau) and the Centers Disease Control and Prevention (CDC) make both TheDataWeb and DataFerrett available to the public without charge. 

This “Mortality” dataset contains geographic, demographic, and cause-of-death variables obtained from the death certificates of people who died in 1998.  Geographic variables include: county and state of residence, and county and state population.  Cause-of-death-related variables include the underlying-cause-of-death coded using the International Classification of Diseases (ICD) Code (9^th Revision).

The coding of death certificate information is standardized across all states.  Death certificates are completed and filed at the state-level.  (CDC, 2005b).  The death certificate information is collected from the states at the federal level by the National Center for Health Statistics (NCHS) and published along with other vital statistics as part of the National Vital Statistics System, “the oldest and most successful example of inter-governmental data sharing in Public Health and the shared relationships, standards, and procedures form the mechanism by which NCHS collects and disseminates the Nation's official vital statistics.” (CDC, 2005d, Introduction section).  “The vital statistics general mortality data are a fundamental source of demographic, geographic, and cause-of-death information.  This is one of the few sources of comparable health-related data for small geographic areas and a long time period in the United States.” (Census Bureau, 2005c, National Center for Health Statistics section).

DataFerrett returns information from TheDataWeb in aggregate form only.  Upon submitting a DataFerrett query for data the following use restriction statement is displayed:

WARNING! DATA USE RESTRICTIONS.  Read Carefully Before Using

The Public Health Service Act (Section 308 (d) ) provides that the data collected by the National Center for Health Statistics (NCHS), Centers for Disease Control and Prevention (CDC), may be used only for the purpose of health statistical reporting and analysis.  Any effort to determine the identity of any reported case is prohibited by this law.  NCHS does all it can to ensure that the identity of data subjects cannot be disclosed.  All direct identifiers, as well as any characteristics that might lead to identifications, are omitted from the dataset.  Any intentional identification or disclosure of a person or establishment violates the assurances of confidentiality given to the providers of the information.  Therefore, users will:

●        Use the data in this dataset for statistical reporting and analysis only.

●        Make no use of the identity of any person or establishment discovered inadvertently and advise the Director, NCHS, of any such discovery.

●        Not link this dataset with individually identifiable data from other NCHS or non-NCHS datasets.

By using the data you signify your agreement to comply with the above-stated statutorily based requirements.

Because DataFerrett queries use the ICD (9^th Revision; ICD-9) codes as a selection criteria, the appropriate ICD-9 codes for each disease were determined through review of an online version of this document available from the National Center for Health Statistics (NCHS, 2005).  See Table 1 for a list of the ICD-9 codes used as selection criteria.  The Disease/Condition DataFerrett Selection Codes were then used to extract the state of residence for those who died in the United States in 1998 from each of the five diseases/conditions of interest.  Data was obtained for each of the fifty (50) states and the District of Columbia (total N for the state-level analyses = 51).  This data was downloaded into an Excel file. 

Added to this Excel file was the population of each state according to both the 1990 Census and the 2000 Census obtained from the Census Bureau American FactFinder, Population Finder website/webtool (Census Bureau, n.d.).  The total 1990 population from the Census Bureau and the total 1998 deaths from DataFerrett for each state were used to calculate the incidence variables used in the analyses.  See Table 2.  The completed Excel file was opened and saved in SPSS (SPSS, 2003), which was used to calculate the descriptive and inferential statistics.  The SPSS file was saved as a Dbase IV file and then opened and saved in ArcGIS for the cartographic analyses.

The same general method was used to obtain data at the county level.  However, in order to protect the privacy of individuals, DataFerrett does not return data for counties with less than 100,000 people according to the 1990 Census.  Instead, all death data for a state from counties with less than 100,000 is lumped into one value. 

Wyoming, for example, has no counties with a population of more than 100,000 so the county-level death data for Wyoming is returned as one statewide number. 

 

Table 1.  ICD-9 Code Used as the DataFerret Selection Criteria and Reasoning

 

Table 2.  Calculation of Variables Used in the Dataset of Variables for Data Analysis

Delaware’s three counties each have a population over 100,000 so county-level data is returned for all three Delaware counties.  New Jersey has twenty-one counties, but three of these counties have a population less than 100,000.  For New Jersey, data is returned for each of eighteen individual counties and then one number is returned for the three counties (combined) with a population of less than 100,000. 

There are 3141 counties in the United States, but DataFerrett returns data on 504, which includes the combined values for a state’s less-than-100,000 counties.  At the county-level, the population data was obtained from Census data available through the University of Virginia (n.d.).  County-level analyses were also done using only those states generally considered to have a high Lyme incidence (Lyme-State).  These 123 Lyme-State counties, which include those counties lumped together because of a less-than-100,000 population, are in the following ten states:  Connecticut, Delaware, Maine, Maryland, Massachusetts, New Hampshire, New Jersey, New York, Pennsylvania, and Vermont. 

All statistical calculations were done using SPSS.  Counts of disease deaths provided by the CDC were normalized by the 1990 Census population information, yielding number of deaths due to a certain disease per 100,000 people in that state or county.  See Table 2.  But normalizing disease deaths by the number of living people in a state or county produced the confounding factor of that geographic unit’s demographics and age.  So a new measurement was introduced: the number of deaths from each disease was divided over the total deaths of each county or state (incidence of death due to a specific disease per 100,000 deaths in that geographic unit).  See Table 2.  Another confounding factor was the exclusion of counties with fewer than 100,000 residents due to CDC privacy policy.  To accommodate for this, the total deaths from all of these smaller counties was smeared proportionally across each county included in the set.  This set of all the counties with fewer than 100,000 people was labeled a “super-county”.  The analysis could use these blocks in combination or independently.

To this data, in both the state and county files, was added the number of new Lyme cases reported each year from 1992-1998, centroid latitude, centroid longitude, and population elevation (the elevation of the county seat or the nearest population center to the county seat for which there is elevation data).  Centroid latitude and longitude were averaged over all counties in a state to calculate the state value.  The same method was used to calculate each state’s population elevation.  Centroid latitude, centroid longitude, and most population elevation information were obtained from the United States Geological Survey (USGS, n.d.).  The Lyme case data was added because the death data from DataFerrett includes more than Lyme (See Table 1).  The DataFerrett category that includes Lyme deaths is “Other Specified Arthropod Borne Diseases” in ICD-9.  This category variable is named OSABD in this study.

The number of Lyme cases in each state for the years 1992-1998 is available from CDC publications (CDC, 2002).  The number of Lyme cases per year by county is not, however, available from the CDC.  Although the CDC publishes some multi-year cartographic material by county, the CDC does not report county-level, annual numerical data for a state to the public.  County-level Lyme incidence data is only available to the public by contacting each state’s department of health, specifically, the state epidemiologist.  In this study, Lyme data available by county was subsequently compiled to match the super-counties data available for DataFerrett death data. 

The process of obtaining Lyme incidence data by county for the years 1992 through and including 1998 was labor-intensive.  Each state’s Department of Health website was visited to see if the needed Lyme data was available on the website.  If the data was not available, that state’s epidemiologist was emailed using contact information from the Council of State and Territorial Epidemiologists (n.d.) website provided by the CDC.  Most epidemiologists contacted via email responded and provided the necessary data.  All of these sources were recorded and the data compiled and added to the database.  As of this writing, this appears to be the most comprehensive database of Lyme in existence. 

Results

            Descriptive statistics for the variables in each of the three basic datasets can be found in Table 3, Table 4, and Table 5.  As many statistical tests assume that the data are normally distributed, each variable’s skewness and kurtosis values and standard errors were examined.  A normally distributed variable has a value of 0 for both skewness (a measure of symmetry) and kurtosis (a measure of clustering around a central point).  If the ratio of the skewness value to its standard error is between –2 and +2, then the distribution is symmetrical (normal).  If the ratio of the kurtosis value to its standard error is between –2 and +2, then the data are normally distributed. (SPSS, 2003; Norusis, 2003).  

Few of the variables are normally distributed.  In the State-Level variables, only MS Death Incidence per 100,000 Live (1990), MS Death Incidence per 100,000 Deaths (1998), Motor Neuron Death Incidence per 100,000 Live (1990), Motor Neuron Death Incidence per 100,000 Deaths (1998), and External Cause Death Incidence per 100,000 Live (1990) are normally distributed.  In the Lyme-State County Level (Population >= 100,000) variables, only MS Death Incidence per 100,000 Live (1990) and Breast Cancer Death Incidence per 100,000 Deaths (1998) are normally distributed.

            The next step in the analysis was a correlation analysis.  Calculating a Pearson correlation coefficient (r) is appropriate for variables that are normally distributed.  (SPSS, 2003, page 379).  Calculating a Kendall’s tau-b or Spearman’s rho is appropriate when the data are not normally distributed.  Because all three of these correlation analyses assume a linear relationship between the variables, a scatterplot graph was constructed for each pair of variables to be analyzed.  Each scatterplot was linear so a Pearson’s, Kendall’s, or Spearman’s coefficient was calculated as appropriate for pairs of variables in each of the three datasets.  The results can be seen in Table 6, Table 7, and Table 8.

Multiple regression was also used to find the model that would best predict the MS Death Incidence per 100,000 Deaths.  All variables contained in the dataset were entered into the regression analysis using the stepwise feature.  All variable values were converted to z-scores for use in the regression analysis.  These results can be seen in Table 9.  Lastly, cartographic analyses were completed.  These can be seen in Figure 1, Figure 2, and Figure 3.  They show the normalized distribution of MS Deaths, OSABD Deaths, and External Causes Deaths, respectively.