WO2001001305A1

WO2001001305A1 - Method and system for accessing medical data

Info

Publication number: WO2001001305A1
Application number: PCT/US2000/010727
Authority: WO
Inventors: Frank L. Madarasz; Ramarao Inguva; James K. Wyly; Joseph Milelli; David P. Krivoshik
Original assignee: International Diagnostic Technology, Inc.
Priority date: 1999-06-25
Filing date: 2000-04-20
Publication date: 2001-01-04
Also published as: AU4251000A

Abstract

A system for accessing medical data comprises a computer system (figure 6, item 608, item 610) providing access to a collection of clinical medical data (figure 6, item 610) and a means for querying the collection of clinical medical data (figure 6, item 608) to determine a diagnosis and probability of successful diagnosis for a patient based upon assessment of the patient to obtain medical data.

Description

METHOD AND SYSTEM FOR ACCESSING MEDICAL DATA

CROSS REFERENCES TO RELATED APPLICATIONS

This application is a continuation in part of United States Patent

Application Serial No. 09/495,185, entitled "Method And System For Accessing

Medical Data" filed on February 1, 2000, and claims the benefit of United States

Provisional Application Serial No. 60/117,509, filed on January 28, 1999 and

United States Provisional Application Serial No. 60/120,309, filed on June 25,

1999.

BACKGROUND OF THE INVENTION

With the enormous amount of information medical information, making a

decision on a factual basis is difficult. The collective experience of clinical

practice forms the basis of virtually all health care decision making. Until now a

readily accessible compilation of collective medical wisdom simply did not exist:

virtually all information, which found its way to efficient dissemination, was of the

research-results kind. While absolutely essential to support medical progress, this

kind of data is essentially useless in the day to day practice of optimum medicine.

There is a need to provide ready access to large amounts of medical data

either formerly unavailable or available in ways that were so cumbersome, or took

so long, as to prohibit any real utility. This ready access can support the query

prediction, diagnoses identification, etc., of the treating clinician. SUMMARY OF THE INVENTION

The present invention is a system for accessing medical data comprises a

computer system providing access to a collection of clinical medical data and a

means for querying the collection of clinical medical data to determine a diagnosis

and probability of successful diagnosis for a patient based upon assessment of the

patient to obtain medical data. A method is also described.

BRIEF DESCRIPTION OF THE DRAWINGS

A more complete understanding of the present invention may be obtained

from consideration of the following description in conjunction with the drawings in

which:

FIG. 1 is a diagrammatic representation of the database architecture;

FIG. 2 is a graphical representation of probability distribution for a

category of illnesses versus the range of illnesses in that category conditioned on

only symptoms and any prior information, P( Ψ | S I ) vs. Ψ;

FIG. 3 is a graphical representation of the most probable distribution of

illnesses conditioned on physical examination, laboratory tests, diagoses ruled out,

patient's medical history and any prior information, P( D | Ψ S' I ) vs. Ψ;

FIG. 4 is a graphical representation of probability distribution for various

treatments conditioned only on the determined set of symptoms, the most probable

illness (see Fig. 2.) and any other prior information P( T | S_f Ψ_p I ) vs. T; FIG. 5 is a graphical of the most probable distribution of treatment

conditioned on additional information, d, such as cost and insurance coverage,

patient's preferences, additional information concerning the patient his medical

history and medication reactions if not included in the prior information, I, P( T | d

S_f Ψ_p I ) vs. T; and,

FIG. 6 is a high level view of a representative implementation of the SMDB

system.

DETAILED DESCRIPTION OF VARIOUS ILLUSTRATIVE EMBODIMENTS

While the present invention method and system for accessing medical data

is particularly well suited for use with human patients and shall be so described, it

is equally well suited for use with other species. The present invention method and

system for accessing medical data offers significant advantages in the practice of

veterinary medicine as well as is generally applicable to treatment of any living

organism. Veterinary medical data can provide veterinarians as well as breeders

with a powerful tool in improving the health of particular animals as well as the

over all breed.

Clinical experience forms the proper basis to inform all decisions in health

care, but it must be collected and carefully monitored. While the financial, social,

and ethical issues must all be taken into account, these are areas already tracked

and the subjects of lively discussion. If, however, a physician, faced with a

presenting patient, wishes to know what the statistical experience of the last, say, 1000 patients so presenting has been, he is lost at sea. If that same physician,

having made a well founded diagnosis, wishes to know the treatment choices

made, and their relative outcomes of the last 1000 or so similarly diagnosed

patients, he is similarly lost. But it is precisely these kinds of data on which the

best medical decisions must be made. The development of the Smart Medical

DataBase ("SMDB") provides a resolution of this dilemma.

The SMDB will provide ready access to huge amounts of data either

formerly unavailable, or available in ways that were so cumbersome, or took so

long, as to prohibit any real utility. The very first advantages are the availability of

the data itself, and quick access in a clinically meaningful manner, made possible

by sophisticated sorting algorithms.

After an initial period, patterns of use, and feedback from users, will inform

future SMDB expansions and modifications to improve its utility. For example,

often-requested series of questions about a disease complex could be placed under

a single command. In addition to such changes, adaptive responses of the system

become part of the system itself.

At every stage the data that is used, and how it is used, is solely at the

discretion of the treating physician. The longstanding and widespread failure of so

called expert systems until now, such as those which read EEG's and EKG's, are

essentially sophisticated sorting programs, simply matching values of

multidimensional parameters with suggested diagnoses. A strength of the SMDB system is that when more sophisticated

"intelligent" functions are introduced, they are already in the statistical test bed in

which they may be evaluated for effectiveness. A new drug is evaluated by clinical

trials, undertaken after a very small research sampling of effect. The clinical trials

themselves, while wider in scope, still remain small relative to eventual field use.

After long-term use in the field, a much larger base of data becomes available

about efficacy, side effects, etc.

The SMDB however, in any of its evolved stages, can be said to be self-

testing without clinical risk. As data on the reliability and accuracy of a certain

SMDB function accumulate, then it become increasingly relied upon by the user

community. The ever-expanding database is always available in real time and

constantly up to date.

The decision-making capabilities of the Medical Protocol and its associated

database structure, to be described in detail below, come under the general heading

of AL What is proposed for the SMDB can be classified under specific

components of AI: Logical AI, Inference, and Learning from Experience. Learning

from Experience is the area of "smartness" in the SMDB. The tools most

commonly employed to implement this component are connectionism, neural

networks, semiotics, and fuzzy logic. The present invention add the use of

Bayesian Statistics. The Bayesian Method specifically focuses on a

predictive/learning capability. Moreover, its statistics are very well developed and

mature compared to the tools presently being used in AI. The present invention is not limited exclusively to the Bayesian Method for the Learning from Experience

component of the SMDB, but consider all viable tools used in AI and those to be

developed. However, it is a powerful method, uniquely different from the tools

presently employed in AI.

While clinicians, the care providers, are a primary target user of the present

invention, insurers and HMO's, the payers, can add their already formidable

financial data to the relevant subset of corresponding clinical information and

achieve a truly effective tool for clinically informed business management.

Similarly, the managers of organizations, as well as public policy makers, can use

the same comprehensive database to make policy decisions genuinely grounded in

reality.

The following symbols and abbreviations are used in this application:

C/C - Chief Complaint C/O - Complaining Of

Dx - Diagnosis

{ ... } - Set, or Collection of, ... ; as in {Dx}

F/U - Follow Up

Glu - Glucose Hx - History

NAD - No Apparent Distress

PE - Physical Exam

PT - Patient

R/O - Rule Out Sx - Symptom — Something patient brings you

Sign - Something elicited from patient by you

URI - Upper Respiratory Infection

WBL - White Blood Cell count WNL - With Normal Limits

Referring to Fig. 1: there is shown the static organizational structure

(architecture) of an embodiment of the SMDB. The SMDB has three major

divisions. The two represented by circles, the Diagnostic segment 102 and

Treatment segment 104 are primarily clinical in origin. The triangular portion 106

consists of that data driven by and collected about costs and payments. These

distinctions reflect the major source of the data in each segment: the multiple

overlapping diagrams indicates that whatever the source, portions of each division

will ultimately be of interest to everyone. The payer is interested in relative

treatment costs and their relation to efficacy, while the MD provider must

increasingly be aware of costs. For instance, the solid circle 108 represents that

small portion of data from all three sources to which the patient has direct access.

The striped portion 110 might be a similar portion accessible to the payer: basic

clinical and cost information without compromising patient confidentiality. There

are as many configurations as there are users.

In a more succinct mathematical approach, the SMDB is comprised of three

major parts: The D (Diagnostic) part 102; the T (Treatment — Outcome) part 104;

the M (MIS — Management Information Systems) part 106. The Venn diagram

picture shown in FIG. 1 is one where all three parts overlap, resulting in seven (7)

distinct pieces: the D, T, M, DT, DM, TM, and DTM.

The D part 102 subdivides into the three-tuple of signs, symptoms, test

results (SST) mapped into the diagnostic set Dx — this is most easily and productively seen as the ICD codes. (The symbol "D" represents one of three main

parts of the database itself, while the "Dx" represents a particular set, the set of

individual diagnoses, within that bigger part D.) The ICD is a worldwide

classification scheme, mostly descriptive but etiologic when it can be, that is used

for almost everything, including billing. The world uses ICD-10, while the USA

alone is still using ICD-9. Strictly, the mapping into is incorrect, since there are

three-tuple images without images in Dx, and members of Dx without inverse

images in the SST. D:Dx:(S,S,T)→Dx

The T part 104 contains maps that take Dx (and also separately a part of

SST) into the treatment set, Tx, (The same kind of relationship holds for the T and

Tx distinction as for the D and Dx sets above: T is a major segment of the data

base, and it contains, amongst many other things, the set of treatment protocols,

{Tx}) and then the T set of these maps to an Outcome set. The Outcome set is an

n-dimensional array with clinically relevant measures which may be in terms

unique to the Outcome set, but which may contain also members of the SST set:

e.g., normalization of white count; five year survival; return to pain-free state;

resolution of fever; normalization of vital signs, etc. The direct SST to treatment

maps describe symptomatic treatment with no ICD identity, admittedly a

somewhat rare occurrence. T:Dx- Tx— ^Outcome

The M part 106 of course has members both in the D and T sets (but not

all), while also having elements unique to itself. These latter contain all the usual

tools and measures typical of MIS systems, and a few unique to the SMDB. The idea associated with the seven pieces described above is being able to

have selective availability. For the majority of physicians, access to part D 102

and part T 104 will be dominant, while those managing their own practices might

have limited availability to part M 106. The harmonious union of clinical

management plus financial information comes from the fact that the billing aspect

of practice is where the two come together. An HMO or other insurer will be most

interested in the M section 106. Hospitals will take an interest in all three pieces,

and it is anticipated that most of Quality Assurance and Utilization Review (QA

and UR) functions will be subsumed and much more efficiently accomplished

using the SMDB organization. Notice that the M part 106 may also contain, in

addition to data, sophisticated tools now used by the industry.

It is anticipated that most of what is needed for all three parts is already

collected under the aegis of clinical record keeping and the billing function. Any

necessary changes will be almost resource-invisible in nature, mainly consisting in

changes to medical forms (including electronic), and how data is organized and

made selectively available (software changes).

A brief example follows to outline the basic aspects of the Bayesian

approach. The number for the probability for the disease is derived using standard

statistical tools as well as Bayes theorem. Let H be the hypothesis whose validity

is being examined. (Please note that in what follows all boldfaced symbols imply

vector quantities, i.e. a possible n-tuple of symptoms, tests, diseases, etc.) In other

words, H(x): The symptoms and test results imply disease(s) x.

Since the set of symptoms and test results are never exhaustive, the validity

of H in general has to be statistical based on evidence. The confidence in the

validity of H(x) is summarized by the quantity P(H(x)|DI) defined by

P(H(x)|DI) = Probability Density(confidence) that H(x) is

true.

Conditioned on the evidence D and prior information I,

D: Set of all symptoms and test results,

and as usual 0< P(H(x)|DI) < 1.

P(H(x)|DI) =1 implies 100% confidence in H(x)

P(H(x)|DI) = 0 implies H(x) is ruled out

The estimation of P(H(x)|DI) is done using Bayes theorem:

P(H(x)|DI) = P(D|H(x)I) P(H(x)|I) / P(D|I)

where

P(H(x)|I) = Prior probability density that H(x) is true based on

previous symptoms and tests on patients other than the one currently under

examination,

P(D|H(x)I) = Likelihood of H(x) based purely on current evidence.

Confidence based on current symptoms and tests,

and

P(D 1 1) = Probability Density for D. The medical database system offers an estimation of P(H(x)|I) using all the

information available in its database. It will also estimate the likelihood based on

current evidence. Based on the database it can also recommend what symptoms to

look for and what tests to order based on previous case histories.

The query responses are a function of the number of symptoms that

matched with the list of known symptoms for a particular disease state, the history

of the patient based on this match of the symptoms, and commonality characteristic

with symptoms in other illnesses. Unless requested this response uses the entire

sample of patient data on the database. The form also has options to choose

specifics that would restrict the sample by cross matching to selected demographic

and medical variables. The software to choose a sub-sample of the entire data set

and come up with a new real time response would then use this later option.

The software would also query the doctor if he wants to see an additional

list of symptoms that might increase the probability of a particular diagnosis. To

confirm the diagnosis the doctor in consultation with the patient might look for

additional symptoms over a stipulated time period, or order tests that might reveal

additional symptoms. He would then repeat, if necessary, the above procedure to

come to a firm decision as to the nature of the illness. At this stage we assume that

the doctor in consultation with the patient and through his software and additional

tests, has come to a decision regarding the illness.

Regarding the choice of treatments, the system presents the option of

running the treatments options menu. Depending on the menu choices, the software gives details about the samples of patients and the response to a particular

treatment as a function of time. It will present a number of statistical processing

results depending on the queries posed up by the doctor (chosen out of a list or

entered by the doctor) such as probability of cure versus probability of failure of

treatment. This is done interactively as a function of time. The software will give

a set of alternatives as a guide to proceed. It will also provide a list of contacts for

consultation should it prove necessary. It will also search additional databases

should further processing be necessary. If the doctor incorporates the patient

treatment and response into the software, it can in real time compare/match with

similar patient scenarios and give useful pieces of information such as

recommendations for midcourse change of medicines/treatment etc.,

The people authorized by the patient and doctor such as the on-call

physician will put the information on a web site for use. The doctor using the same

software can use a utility to post a question on the bulletin board, protecting the

patient's identity, to which other people on the network can respond.

SMDB updates occur in real time, allowing analysis and decision making at

the time and place of physician contact. In addition, correlations can be had

sorting for patients with constellations of illnesses and conditions, and all the

relevant probabilities of diagnoses, treatments, and outcomes derived for any

specialized situation. In this way the vast amount of accumulated clinical data may

be queried for the maximum it can inform about a particular patient. Two functions of the SMDB which will have increasing importance in the

future will be the telemetering of patient data via phone/cable lines to physicians'

offices and hospitals, and the transmission of lab results and imaging data for the

best expert interpretation. The former will be of use in both crisis management

(e.g. asthma attacks) and maintenance (e.g. blood pressure, glucose, or pulmonary

function monitoring).

Further along in a case, the SMDB will allow an informed analysis of failed

treatments and treatment alternatives.

Lastly, such a massive integration of clinical data will allow relatively

effortless retrospective studies of diagnostic accuracy, treatment efficacy, and

safety. It will permit, within confidentiality legislation and guidelines, the

convenient access of individual patient histories and data by those people and

agencies entitled to that data.

The Bayesian Statistical Approach

The Bayesian approach gives the posterior probability density as a function

of the prior density functions, while a utility function is used to order preferences.

These things are defined carefully in terms of our envisioned structure and possible

queries. Decisions relying on the Bayesian approach use the following:

A space of possible actions by the decider - the clinician.

• A space of possible patient states, diagnoses for example.

• A collection of possible experiments, or tests, or findings.

• A space of possible outcomes for these tests, experiments. • A utility function defined over all the possible consequences which defines

outcome preferences.

• The above are then used to "condition" the prior probability density

function as more results (information) become available, yielding the posterior

probability.

While SMDB can be implemented with a centrally located data base

accessed over a computer network such as the internet using, utilizing established

communication protocols and browser type interfaces, the present invention is also

equally well suited for a distributed architecture implementation.

The effective data compression provided by Bayesian Statistics lends itself

to a distributed architecture implementation. For example, millions of patient

records are reduced to sets of conditional probabilities, which only require 4-digit

resolution to be effective. Consider the following: a billion patient records can be

distilled into a table containing, perhaps 10,000 sets of conditional probabilities.

As the size of the database increases, the table values may change; however, the

size of the table is unchanged.

In one embodiment of the SMDB, the collection of patient records is hosted

on a network accessible, centralized system. Each user is provided with software

enabled to run on an individual computer or over a local network. The software

consists of the user interfaces; the algorithms described herein and the tabular

database of conditional probabilities and interactive diagnostic query responses.

Thus, the physician will be able to utilize the database in an efficient mode. There are several methods that can be used to update the database. One

method employs "push" type technology, wherein the central database sends, daily

updates of the tables as well as other relevant information to the user. Another

method employs downloading updates, such as when a patient's records are

uploaded to the SMDB central database. Either method is efficient as the updates

would be minimal in volume and typically only require a few seconds to transfer.

However, the simpler option of accessing the centralized SMDB, with a

simple browser, and performing the patient data upload and diagnostic

query/response mode is useful when physicians are away from an office

environment and can access the system using cell phone, personal digital assistant,

such as a palm pilot, or lap top computers.

The following is an simple exemplary embodiment of the SMDB and the

application of Bayesian Methodology to a Clinical Case.

Statement of Illness and Background Information on Patient

Mr. Sam Jones is a 37-year-old married white male who presents with

multiple upper respiratory complaints. He is new to this clinic secondary to his

having recently changed employers, and hence insurance carriers. He currently

works as a janitor at a local school. Chief Complaint: T have this cough, and my

throat is very sore.' Complains of night sweats, waking the patient two to three

times a night for the last three nights. Complains of cough, non-productive at first,

now with mucopurulent discharge, occasionally blood tinged, times five weeks.

Construct Initial Probability Distribution

P( Ψ | S I ) =

Conditional Probability for category of Illness - Ψ (Vector)

Conditioned on Symptoms Vector S and Any Other Prior Information

Concerning the Patient and Their Medical History

Any illness Ψ,, Ψ₂, Ψ₃, ... c Ψ and Ψ_p is likely. The Ψ.'s are a range of illnesses

that correspond to the stated symptoms given in S, and Ψ_p is the final outcome — a

specific illness with an associated probability of being true — of the analysis. The

result of this construct would typically look as that displayed in Fig. 2.

Medical Examination

Physical Examination

1. Vital signs

a. Blood pressure: 124/70

b. Pulse: 85

c. Respiration: 20 d. Temperature: 101 degrees Fahrenheit

2. General description: the patient is a 37 year old married white male who looks

his stated age; he is pleasant, appears well nourished, and seems in an overall

good state of health.

3. Skin: warm and dry; turgor adequate; color normal. There is no icterus,

purpura, rash, or unusual pigmentation noted. Hair normal in appearance,

distribution, texture.

4. Lymph nodes: no cervical, supraclavicular, axillary, epitrochlear, or inguinal

adenopathy.

5. HEENT (Head, Ears, Eyes, Nose, Throat):

a. Head: normocephalic and atraumatic; no lesions noted.

b. Eyes: cornea without lesions, conjunctiva clear, sclera white. Pupils

are equal, 3mm in diameter, round, reactive to light and accomodation.

Extraocular movements within normal limits without nystagmus or

strabismus. Fundii are benign. Disks well delineated. There are no

hemorrhages or exudates. Visual acuity is 20/20 bilaterally, and visual

fields are within normal limits to confrontation.

c. Ears: normal in appearance. Auditory canals clean and without lesions.

Tympanic membranes intact. Hearing adequate.

d. Nose: septum within normal limits and without deviation. Nasal

mucosa pink and without abnormal discharge. No nasal polyps or

other lesions. Frontal and maxillary sinuses nontender. e. Mouth and throat: lips without cyanosis or pallor. Buccal mucosa

normal in appearance. Teeth in good condition. Tongue without

lesions or tremor, protrudes midline. Pharyngeal mucosa is

erythematous and without other lesions, exudates, or evidence if

inflammation. Gag reflex intact.

6. Neck: neck is supple, full range of motion. No evidence of tracheal deviation,

jugular venous distension, or lymphadenopathy. Carotid pulses are 2+, equal

bilaterally, and without bruits. Carotid upstroke is within normal limits.

Thyroid normal in size, palpation reveals no nodules or masses.

7. Back: spinal curvature is normal; no scoliosis, kyphosis, or tenderness. Full

range of motion present.

8. Chest: thorax is symmetric. Full expansion bilaterally. AP diameter is within

normal limits.

9. Lungs: fremitus is equal bilaterally. Lung fields resonant throughout. Breath

sounds and voice sounds normal. There are no rales or ronchi, but some end-

expiratory wheezes throughout, more prominent in the bases bilaterally.

10. Heart: palpation reveals no heaves or thrills. The PMI (point of maximum

impulse) is medial to the midclavicular line, fourth intercostal space.

Auscultation reveals SI, S2, of normal intensity. There are no S3, S4, rubs,

clicks, or other abnormal heart sounds. Heart rate is 70 BPM and rhythm is

regular. 11. Breasts: breasts are symmetric and of normal contour. Skin is of normal color

and appearance; there is no edema, ulceration or erythema. Nipples are of

normal size and shape; there is no nipple retraction, ulceration or discharge.

Palpation does not reveal any tenderness or masses.

12. Abdomen: normal size and contour. There are no capillary dilatations, skin

lesions, or surgical scars. Auscultation reveals normative bowel sounds and no

abdominal bruits. Palpation reveals no abdominal tenderness, guarding, or

masses. Liver edge is felt approximately 1 inch below the right coastal margin;

it is firm, sharp, and smooth. The liver percuses to approximately 8 to 10 cm.

total span. The spleen is not palpable.

13. Rectal exam: no external anal lesions. Sphincter tone normal. No internal or

external hemorrhoids. Rectal mucosa appears normal, with no nodules or

masses present. Stool is brown and negative for occult blood.

14. Genitalia: inspection reveals normal distribution of pubic hair. No lesions or

discharges. No external lesions. Testes are descended, nontender, of normal

size, without nodules or masses.

15. Inguinal area: no lymphadenopathy noted. Femoral pulses are 2+ and equal

bilaterally. Auscultation reveals no femoral bruits.

16. Extremities: there is no clubbing, cyanosis, or edema. Brachial, radial,

popliteal, dorsalis pedis, and posterior tabialis pulses are 2+ and equal

bilaterally. Musculo skeletal exam reveals no joint deformities and full range of

motion. There is no bone, joint, or muscle tenderness noted. 17. Neurologic: patient is alert and oriented to time, person, and place. Cranial

nerves II to XII are within normal limits. Speech, memory, and expression are

within normal limits. Muscle strength is 5/5 in both upper and lower

extremities. There is no muscle atrophy or involuntary movement noted.

Testing of cerebella function reveals normal gait, negative Romberg test, and

good coordination in finger-to-nose, heel-to-shin, and alternate motion testing.

Sensory is intact to light touch, pain, and vibratory stimuli. There are no focal

motor/sensory deficits. Deep tendon reflexes are 2+ and equal bilaterally.

Chosen Tests Include:

1. Throat culture and sensitivity (including strep): rapid strep test (poor but

widely used) is negative. 24 and 48 hour cultures are positive for

streptococcus sensitive to a wide range of older antibiotics. Mycoplasma

culture will take a week or more to return, somewhat more difficult to do

reliably, and is not chosen.

2. Chest X-ray, PA and LAT (back to front and side views). The pictures

show very light patchy bilateral infiltrates, without consolidation.

3. A PPD (TB test) is chosen, and planted. A mycobacteria culture is not

chosen (possibly a mistake) at this time.

The Diagnoses to be Ruled Out Include:

1. Strep infection

2. Pneumonia

3. Residual mycoplasma infection 4. Mycobacteria infection (tuberculosis)

Construct Data Set (Vector) D and New Symptoms Vector S'

The data set is made up of the information contained in the seventeen (17) steps of

the Physical Examination, that obtained from the three (3) Laboratory Tests and

from the four (4) Diagnoses Ruled Out:

D = {(D,, D₂, ... D₁₇ — Physical Exam), (D₁₈ D₁₉, D₂₀ — Lab Tests),

(D₂₁ D₂₂, D₂₃, D₂₄— Ruled Out)}

In addition, the physician and/or the patient may have identified one or more

symptoms not initially reported. The symptoms vector must be updated:

S' = S + Additional Symptoms Observed.

Construct a New Probability Distribution for the Illness Ψ Using D, S' and Bayes' Theorem

P( Ψ | D S' I ) =

Conditional Probability for Reduced Range of Illnesses - Ψ (Vector) Conditioned on New Data D, Updated Symptoms Vector S' and Any Other Prior

Information Concerning the Patient and Their Medical History

P( Ψ I D S' I ) = [P( D I Ψ S' I ) P( ψ I S' I )] / P( D | S' I ),

where

P( Ψ I S' I ) = Prior probability density that Ψ is true based on

previous symptoms and tests on patients other than the one currently under

examination, P( D I Ψ S' I ) = Likelihood of Ψ based purely on current evidence.

Confidence based on current symptoms and tests,

and

P( D I S' I ) = Probability Density for D.

The new distribution is much narrower in the range of illnesses it spans and peaks

about the most probable illness, Ψ_p as illustrated in Fig. 3. At this point, the

physician, based on other past case experiences with similar symptoms and/or

professional instinct, has an option of selecting the most probable illness, Ψ_p, to

treat or one of a small number of other illnesses within a nominal 0.99 probability

"threshold" as indicated in Fig. 3.

A Presumptive Diagnoses of possible strep infection, possible mycoplasma

infection is made. Less likely is TB. The patient will be treated as an outpatient.

The SMDB is accessed for presenting symptoms and signs. It is found that

in the last 30 days there have been 19 cases of students or staff of that school

diagnosed with mycoplasma infection. This makes the diagnoses now

overwhelmingly likely.

All have responded to a course of erythromycin (not universally true for

this infection). Inhaled bronchodilators, and inhales steroids.

The Patient Treatment Plan

The SMDB is designed to be an effective tool in helping to select a course

of treatment for the patient. Two in puts are needed: the most probable illness, Ψ ,

and a final determined set (vector) of symptoms, S_f, which may be identical to S'. When this information is put into the SMDB it returns a probability distribution of

treatment, T,

P( T | S_f Ψ_p I ) =

Conditional Probability of Treatment(s), T

Conditioned on Symptoms Vector S_f and Any Other Prior Information I

Concerning the Patient, His Medical History Including Medication

Information

As illustrated in Fig. 4, the return may be a suggested board range of treatments,

T„ T₂, ... .

Tailoring the Treatment Plan

The treatment plan may be tailored to conform to additional information.

First a new data set (vector), d, is constructed. This data set will contain

information such as simplicity of treatment, cost and insurance coverage, patient's

preferences, additional information concerning the patient his medical history and

medication reactions if not included in the prior information, I, etc., etc. From this

information the SMDB will employ the Bayesian Methodology and construct the

most probable distribution of treatments.

P( T I d S_f Ψ_p I ) = [P( d I T S_f Ψ_p I ) P( T I S_f Ψ_p I )] / P( d I S_f I ),

where

P( T I S_f Ψ_p I ) = Prior probability density that T is true based on the

determined illness, Ψ_f, the symptoms, S_f, associated with Ψ_f and other prior

information, I. P( d I T S_f I ) = Likelihood of T based on no directly knowledge of

Ψ .

and

P( d I S_f I ) = Probability Density for d.

The new distribution is much narrower in the range of treatments it spans and

peaks about the most probable illness, T_p as illustrated in Fig. 5. At this point, the

physician, based on other past case experiences with similar situations and/or

professional instinct, has an option of selecting the most probable treatment, T_p, or

one of several other treatments that nominally fall within 0.99 probability

"threshold" as indicated in Fig. 5.

A course of erythromycin (both because of the above information, and to

nail the last doubt about a possible strep infection, and helping to prophylax against

the latter).

• A two week course of an inhaled bronchodilator.

• A two week course of an inhaled steroid.

• Return in two days to have PPD read.

• Return in two weeks for follow up and to schedule an initial intake to the

service.

The PPD is negative in two days: no action taken. Patient reports good

symptomatic relief with the inhaled medication. Two-week follow up shows a satisfactory resolution of symptoms:

productive cough is absent times 8 days; patient is afebrile. Lungs are now clear.

No follow up chest X-ray is needed. Oropharynx is now clear with no erythema.

Update/Upgrade SMDB

This patient's medical event now become part of the SMDB, and hence part

of the knowledge informing subsequent decisions for other uses of the SMDB.

While this case does not involve life and death issues, as do cancer

diagnosis and treatment, it is sufficient to illustrate the utility of the present

invention SMDB. Unless the treating team was directly involved in the other cases

from the school, or had prescient knowledge about them, this confirmatory data

would not be available. While in this case the resulting benefits included higher

confidences in diagnosis and treatment, increased treatment efficacy, lower

resource consumption all around, (all worthwhile goals), in most cases in the

management of serious chronic illness, these advantages will accrue a thousand

fold more. In addition, the issues of quality of life, morbidity and mortality,

resource availability, access to treatment, are all vitally important areas, which will

see significant benefits from the SMDB function.

Moreover, a portion of the SMDB subsumes the typical medical record and

administrative record functions. This alone will not only enhance those roles, but

also reduce costs involved with them.

Referring to Fig. 6 there is shown a representative implementation of the

SMDB system. A personal digital assistant 602, used by the care provider, contains client software and intelligent agents for providing user interfaces and

communicates by a wireless link 604 to a wireless communication device 606. The

remote communication device 606 is coupled to a local computer system or server

608. The local computer system or server 608 accesses the SMDB server 610,

through the Internet, dedicated network, or other suitable communication links.

Numerous modifications and alternative embodiments of the invention will

be apparent to those skilled in the art in view of the foregoing description.

Application to systems of other living organisms as well as electro/mechanical

systems having non specific diagnosis for failure or problems are within the scope

of the invention. Accordingly, this description is to be construed as illustrative

only and is for the purpose of teaching those skilled in the art the best mode of

carrying out the invention. Details of the structure may be varied substantially

without departing from the spirit of the invention and the exclusive use of all

modifications which come within the scope of the appended claim is reserved.

Claims

WHAT IS CLAIMED:

1. A method for accessing medical data comprising the steps of:

assessing a patient to obtain medical data;

accessing a collection of clinical medical data;

querying the collection of clinical medical data; and

determining a diagnosis and corresponding probability of successful

diagnosis for the patient;

wherein said corresponding probability of successful diagnosis is defined

by P(H(x)|DI), probability density that said diagnosis, H(x), is true, said diagnosis

H(x) is a function of symptoms and test results x, D represents evidence and I

represents prior information.

2. The method as recited in claim 1 wherein the step of accessing the

collection of clinical medical data utilizes a computer network.

3. The method as recited in claim 2 wherein said computer network is a global

computer network.

4. The method as recited in claim 1 wherein the collection of clinical medical

data is distributed.

5. The method as recited in claim 1 wherein P(H(x)|DI) is estimated using

Bayes theorem.

6. The method as recited in claim 1 wherein P(H(x)|DI) is estimated as

P(D|H(x)I) P(H(x)|I) / P(D|I), where P(H(x)|I) is prior probability density that

H(x) is true, P(D|H(x)I) is likelihood of H(x) based on current evidence and

P(D 1 1) is probability density for D.

7. The method as recited in claim 1 wherein the patient is a human.

8. The method as recited in claim 1 wherein the patient is an animal.

9. A system for accessing medical data comprising:

a computer system providing access to a collection of clinical medical data;

and

a means for querying the collection of clinical medical data to determine a

diagnosis and corresponding probability of successful diagnosis for a patient based

upon assessment of the patient to obtain medical data;

wherein said corresponding probability of successful diagnosis is defined

H(x) is a function of symptoms and test results x, D represents evidence and I

represents prior information.

10. The system as recited in claim 9 wherein access to the collection of clinical

medical data utilizes a computer network.

11. The system as recited in claim 10 wherein said computer network is a

global computer network.

12. The system as recited in claim 9 wherein the collection of clinical medical

data is distributed.

13. The system as recited in claim 9 wherein P(H(x)|DI) is estimated using

Bayes theorem.

14. The system as recited in claim 9 wherein P(H(x)|DI) is estimated as

H(x) is true, P(D|H(x)I) is likelihood of H(x) based on current evidence and

P(D 1 1) is probability density for D.

15. The system as recited in claim 9 wherein the patient is a human.

16. The system as recited in claim 9 wherein the patient is an animal.

17. The system as recited in claim 9 wherein the computer is a personal digital

assistant.

18. The system as recited in claim 9 wherein the means for querying the

collection of clinical medical data utilizes a wireless communication interface.