What is the best way to define events over space and time?
The best use of geography for epidemiological purposes focuses on adjusted rates. These are actual numbers for a given set of people that may be compared with the expected or the norms. These numbers are calculated such that the population they represent is identical to or close to the population they are being compared with. In the case of U.S. population health, the standard that all populations are compared to is the U.S. population as a whole. In the case of a state or county health assessment, we like to compare our research population to the state and then to the nation.
The purpose of this standardization or normalization of the data is to make whatever comparisons our work results in credible. It would unfair for example to compare the numbers of fractures and injuries in professional sports game players with those in non-professional or school- and college-sponsored sports. Not only is the degree to which the sports activities are engaged in very different, so too are the average ages of the people being evaluated and compared to each other for such things as emergency care needed for dislocations, wrist and ankle sprains and fractures, skull or jawbone fractures, concussions, severe facial injuries, or severe sprains, strains, cuts, bruises and the like. The age factor alone makes these two populations incomparable. But also the gender content, the past medical history, the underlying chronic disease and long term medication and health related complications problems, and so on.
With population health, researchers and statisticians like to compare their particular population with the national standards. These standards may be everyone in the country within the given age, gender, ethnicity grouping. These standards may also be based upon a well controlled sample of the overall population. Because the two populations being compared are most likely not a one-to-one comparison, percentages of total population (adjusted to mimic the research population) are evaluated. This may also be done by analyzing and reporting on prevalences or incidences.
In general, these numbers provide us with something that we feel comfortable with comparing, and basing many of our decisions upon.
A Test Run on compliance and cost analysis
There are two directions we can go with numbers pertaining to population health. We can look at the true values and work with that data to change what’s happening, or we can work with the readjusted scores and then try to make changes and then measure how effective these changes are. These two processes at first sound very similar, but when it comes to infrequent to rare events they aren’t.
The best use of population data requires we use two approaches to the data. We must first apply the standards–the adjusted data for purposes of obtaining our traditional outcomes information. The advantage to the traditional approach is it gives us results that most of us can understand, like the likelihood of having a particular condition is 6 in 1000 or the probability of dying in the next year is 3% for our age group.
But sometimes, percentages mean nothing. They are least valuable when it comes to real life processes, like paying your bills, making sure you can see your doctor, trying to avoid exposure to that other person in your town whose kids are carrying a disease potentially fatal to you or your parents living with you. These discrete events do have a probability, but the percent likelihood that an event is going to happen doesn’t always concern us. We are not at all interested in these percentages, as evidence by simple human behavior–if we were that way the lottery wouldn’t exist, and we wouldn’t risk gambling on certain races. We gamble because the race is happening, and buy the lottery ticket because the opportunity is there. Likewise, we fear catching a disease because we are told that the probability is there as an event, with one of two probabilities–either it impacts us or it doesn’t.
When it comes to money and health, cost is a lot like this problem we have with choosing whether to buy a ticket or not, just like we chose whether to invest in health insurance voluntarily or not. The cost for insurance is so high to many, that you have to have an extremely powerful reason to drive you into such an investment. When betting on horses, you at first do not provide a high stakes wager. You go from low or moderate to high, but once once the adrenalin kicks in and changes the thinking process.
Health/Cost Scores: an early regional metrics trial run
With health care, when we subconsciously pay several hundred to a thousand or more a month, we don’t necessarily look at tit that way. It is taken from our salary, and what we perceive is how much of our salary is take home. If all of this were reversed, and we were asked to voluntarily decide whether to go for a $350 a month option of a $1500 per month options, which one are we most likely to go for if we only had these two options–probably the first, with hopes that the odds don’t work against us.
The same is the case for cost mapping and health, and this is how the businesses want to look at a spatial health data when its comes to balancing your budget.
The problem the population health care statistics poses to business is that people health metrics has one set of rules for evaluating a system, and cost another.
In the case of severe hemophilia, the cost of a week’s worth of medicine can be as high as $100,000. No single person expects to get older and have to depend upon such a costly prescription products.
Yet a few of these people exist in the country, and if you go to a place and find that within that small area, there are 50 such people, if you are the insurer for all of these people, this means their drugs will cost you 5 million dollars per month, or in theory 60 million per year!
Is this even possible? Unfortunately yes. If that hemophilia was genetically linked and culturally linked to unplanned or unexpected third or fourth generation in-breeding, it becomes even more possible. And it is happening.
But there are plenty more medicines out there that are high cost, and which engage more people than the rare hemophilia medicine, and produce much higher costs that the hemophilia medicine that we need to be concerned with.
That is the reason raw data/case related spatial mapping needs to be employed as much as prevalence/incidence adjusted population rates.
There are more financial benefits that can be gained by engaging in this NPHG process with this goal in mind–reducing costs.
Testing N versus Prevalence
If you are now convinced that costs are a concern, let me turn the attention toward medical or diagnostics claims types, and use the same logic. There can be two towns, one with 200 cases of spouse and child abuse per month, and another with just 10. The city with 200 per month has 10,000 families; the city with 10 has 500 families. Both are of the same percentage. Interventions planning may or may not be the same for both of these towns. If the diagnosis was controversial, and even a few cases a major concern, and we change this to a sensitive issue like child sexual abuse, teenage prostitution, severe malnutrition and mental health disease, or infibulation, no matter what the numbers generate for us in terms of percentages, percents really don’t matter with the intervention process. In a city of several million people, with just 100 cases of teen age suicide taking place per year in the form of a ‘Run Amok’ diagnosis, would that mean that due to the very low percentage rates that we ignore such cases? Of course not.
High costs and high social concerns negate the value of statistics in any of these kind of population health studies. The reasons to be able to in a single effort, quantify such statistics for your area to to enable this kind of population health intervention process.
A form of Pica from the Caribbean
Be our focus on money, or on social issues and health, the results are the same. We allow the standard prevalence/incidence work to be engaged in at the population health level for diabetes, rheumatism, hypertension, obesity, osteoporosis, what have you, satisfying the need of the bulk of the population in general, but the population as a society in general doesn’t have all of its needs being met when we ignore things like abuse, criminal behavior, behavioral health diseases, learning disability, suicide and aggression, the way that we do when we focus only on population health reporting at a very broad scale. To make significant changes in the health care system in general right now, we have to be able to map everything, not just those token one or two hundred metrics that exclude some of the sickest people in a population, the kinds of metrics being used right now to evaluate population health.
This is why numbers, even for small populations are important, and is why a mapping of both disease counts and prevalence is required for the best intervention programs to be developed.
The following are cases where this philosophy prevails. The numbers of patients, due to their spikes, are much more important than the percentages. Due to the nature of these case spikes, and their implied meaning and risk, we have to pay heed to these spikes where ever they exist. This is why they are mapped using this 3D modeling tool. This is also why 3D modeling performs some unique functions that the standard 2D modeling techniques can never provide us with.
The conditions or diagnoses labelled as “African diseases” are those with genetic, physiological, cultural and microbiological linkage to Africa. Spatially, the heart of African diseases in the U.S. is primarily related to the slave in-migration routes. Diseases like Sickle Cell demonstrate this. But African culture also had several cultures not as closely tied to the slavery years such as the Muslims. There are a few culturally-bound syndromes and culturally-linked diagnoses related to to these people. The peak on the map above is adjacent to the Great Lakes, and related to diagnoses coming out of the Chicago-Detroit region.
African Diseases (one of several partial, categorical runs)
African Diseases (Another run)
Obscure African Cardiomyopathy
SOCIAL MEDICAL ISSUES
Homeless Teenagers and Young Adults
Homeless 66+ year olds on the street
Child abuse by kids
Shaken Baby Syndrome
Kids engaged in off-road vehicle ATV accidents.
Kids under 5 yo who are struck by cars
Crack baby Syndrome
Takotsubo versus Factitious Disorders (Munchausen Syndrome) in Elders — An oriental vs. American older aged spouse behavioral disorder comparison
So what are we gaining by using NPHG?
The tradition is to rely upon N and Incidence/Prevalence [NPI] maps. For decades this form of mapping has been developed and improved upon for demonstrating final outcomes.
Very few of us would ever question the value of the standard NPI maps.
One of the limits to standard NPI maps is they are two-dimensional. This limits their applications. There are some standards out there for generating 3D versions of these known as semi-cube maps. But these semi-cube maps again have their limitations when it comes to applications in the field and fast problem solving.
If you were an epidemiologist trying to find the source for an outbreak, you would never rely upon a standard NPI map. Even at the small area level, using census blocks to define you stats, or smaller areas, the PI (Prevalence-Incidence) approach does little to provide you with what you might need to know. The limiter here is the way in which that data is presented.
Other problems with standard mapping are the broad ranges in use in the past for national PI data. With population health, we blend different ages into groups, so a ground with fourteen 5-years olds and one 9-year old with a disease, for a total of 14+1 = 15, would seem the same as a group with five each of the 6, 7, and 8 year old children using the older system for mapping data. One-year age groups has to be used to produce reliable data, and 3D demonstrations of the results used to demonstrate small area nearest neighbor differences.
The more gratifying aspect of the results of using NPHG mapping is that its three-dimensional nature and attention to detail paid at the small area level is what makes it so useful. With NPHG we know immediately where the nidus is, and if we want, where the highest PI place is as well (although in my algorithm I refer to something similar to this as IP).
If we take what we see on the 3D model and apply it to the other 2D (landuse ,surface application–a map of places and place name) knowledge that we have, we know exactly where the problem lies. We can use that information to turn to traditional mapping if we want in order to develop an intervention activities process.
Putting this to the test . . .
The following are four ways to present the Lyme Disease problem. These are all reporting standards with three of the four CDC generated.
Lyme Disease (one of several runs)
MAPPING INTERNATIONAL DISEASE ISSUES
Chinese Liver Fluke
Korean Hemorrhagic Fever
For establishing intervention programs, the above doesn’t provide much locally useful data. NPHG maps, even at the national level, point to places to start.
With closer mapping, details about specific regions can be seen and even more comprehensive programs initiated. The obvious impact of topography on the diffusion of lyme disease for example is more visible in up close mapping at different angles of the disease patterns, demonstrating a very steep peak east of the line forming the tops or ridges of the Appalachian and allied mountain chains.
Hudson Valley Lyme Disease
UNITED STATES NATIVE DISEASES
MORE FOREIGN BORN:
Crimean or Congo Fever
FOREIGN DISEASE PENETRATION IN GENERAL (>200 ICDS evaluated as a group)
ASIA (partial review; this was done in two parts; complete set for worldwide continental/regional analyses posted elsewhere)
MIDDLE-SOUTH AMERICA (partial review)