Method of diagnosing or categorizing disorders from biochemical profiles

Abstract

A method for diagnosing disorders in a subject organism, in which fluid samples from normal and abnormal organisms are analyzed to generate electrical signal patterns representative of molecular constituents of the samples. A data base of electrical signal patterns representative of frequency distribution of sample constituents from the abnormal organisms having known categories of disorders and control samples from normal organisms are created, and a fluid sample taken from the subject organism is analyzed by comparing it to the data base for conformity to the electrical signal patterns representative of the frequency distribution. The invention has particular applicability to assisting in the diagnosis of degenerative diseases such as Alzheimer's Disease, Parkinson's Disease, Huntington's Disease, schizophrenia, amyotrophic lateral sclerosis and Progressive Supernuclear Palsy.

Description

This invention relates to analytical and mathematical methods for diagnosing or categorizing disorders. The invention has particular utility for diagnosing or categorizing disorders in living animals from analysis profiles of biologically active materials such as neurotransmitters and other neurochemical substances in brain tissue, cerebrospinal fluid, plasma, serum, saliva, blood containing platelets, nasal mucosa, urine and the like, such as catecholamines, their precursors, cofactors and their metabolites. The invention is uniquely capable of differentiating a large number of compounds of biological, diagnostic and/or pharmaceutical significance and of using such differential for diagnosing disorders and will be described in connection with such utility although other uses are contemplated.

There is an extensive body of literature relating abnormalities in neurotransmitters, precursors, and metabolites to degenerative, neuropsychiatric and behavioral disorders, hypertension and certain carcinomas. See, for example, Schildkraut et al in The Brain, Biochemistry and Behavior , Proceedings of the Sixth, Arnold O. Beckman Conference in Clinical Chemistry, pages 47-68. Although the potential role of these compounds in a number of significant disorders has been established, their routine analysis has not yet achieved widespread clinical use. Two problems in the clinical utility of neurotransmitter measurements are related to the economic and technical limitations of current technology. First, there is felt to be a high degree of interlaboratory and intersample uncertainty in quantitative values. Second, it has been difficult to measure enough of the known metabolically related compounds of a particular neurotransmitter to fully describe its biochemical significance in an individual sample, or to detect, identify and measure unusual neurotransmitters—an important aspect of basic research in various disease states that is presently very expensive and specialized.

While a number of interlaboratory technique intercomparisons for a variety of neurotransmitters have been carried out, there has been no comprehensive study within and among different techniques and laboratories for neurotransmitters in typical samples of interest. In the absence of such studies, given the complexity of the analytical problem and the historically wide variation whenever an analyte has been subjected to rigorous interlaboratory testing, the current values for normal and abnormal neurotransmitter levels must be taken with unspecified and probably wide limits of confidence.

Although the analysis of single neurotransmitters or metabolites from a complex biochemical pathway has been shown to correlate with a number of disorders utilizing statistical analysis over a large number of samples, the analytical level of a single neurotransmitter in an individual sample, with a few exceptions, has had relatively low clinical diagnostic utility. Essentially the state of the field of biochemical correlates of disorders is that while between large populations of normal and abnormal individuals a correlation generally can be determined for a particular biochemical, the scatter that results from both analytical and biochemical phenomena typically does not permit the level of a particular biochemical to be utilized diagnostically for a particular single individual. Nor may a single biochemical value be utilized for the rational prescription or development of a pharmaceutical for that individual. This is not particularly surprising in that both the levels and effects of a particular neurotransmitter are modified by a number of other neurotransmitters, in the same, or parallel metabolic pathways. If, for instance, 5-HT (serotonin) is to be used as a diagnostic tool for depression, suicidal tendencies, or schizophrenia, it would be necessary and perhaps provide a route to definitive diagnosis and pharmaceutical specification or development, to simultaneously determine the approximately 40 other compounds that derive from tryptophan and significantly effect the indolaminergic neuronal system's activity.

In recent years, LCEC (Liquid Chromotography with Electrochemical Detection) has become a common tool for the determination of catecholamines biogenic amines and their metabolites in biological fluids. Because of sensitivity limitations (typically 20-50 pg) and the complexity of biological samples, both separation and concentration steps typically have been necessary. Heretofore, plasma catecholamine analysis typically required three steps. First, the sample is collected and the catecholamines separated and concentrated, for example, using the alumina extraction procedure of Anton and Sayre (See A. H. Anton and D. F. Sayre, J. Pharmacol, Exp. Ther., 138 (1962), p. 360-375). The analytes, norepinephrine, epinephrine and dopamine, along with the internal standard DHBH (dihydroxybenzylamine), then are separated chromatographically, and finally detected electrochemically. Typical sample size requirements are 1.0 ml plasma or serum. In routine clinical use, there have been numerous problems with conventional techniques (alumina absorption, ion exchange and extraction), due to a large number of poorly understood variables, in the overall analysis system of sample acquisition, storage, preparation and sensor response. These problems have quite likely confused the relationships that may exist between levels and distribution of the catecholamines and various physiological and behavioral phenomena and disease states.

In the analysis of complex biological materials such as blood, serum and cerebrospinal fluids which may contain numerous different constituents, the important (e.g. abnormal) metabolites such as neurotransmitters to be identified may be present in only parts per trillion. While a chromatographic column can achieve macro separation of the various constituents, it may not provide adequate spatial (in time) separation of the extremely small portion of metabolites of interest from the much larger percentage of the many other compounds coeluted from the column at the same time as the metabolites of interest. Many of these interfering coeluted materials are electrochemically active but electrochemically irreversible, while many metabolites such as neurotransmitters are both electrochemically active and electrochemically reversible. It has been found that the analytical problems of reliable measurements of neurochemicals and related compounds are complicated by the fact that interferences with conventional or prior technologies are disorder related. This was discussed in my prior publication, (Matson et al, Clinical Chemistry, Vol. 30, No. 9, 1984) (see U.S. Pat. No. 4,511,659) for dopamine, dopac and seratonin measurements in directly analyzed brain extract anti cerebrospinal fluid for normal, schizophrenics and Alzheimers. Recent work has indicated that even for the widely used and accepted technique of alumina extraction for plasma catecholamines that interferences may be disorder specific. Higher values for Norepinephrine (NE) and Epinephrine (EP) were observed following alumina extraction and analysis of a single energy conventional electrochemical detector than for a three cell redox detector on neonatal stress blood samples. Analysis of the neonate extracts on the sixteen channel chemical imaging system revealed several unexpected compounds that are potential interferences including dihydroxyphenylacetic acid (DOPAC), 3 hydroxykynurenamine (3-OHKYA) and 3-hydroxy-anthranilic acid (3-OHAN). These compounds have not been detected in normal adult plasma alumina extracts.

In my aforesaid U.S. Pat. No. 4,511,659, there is provided an electrochemical detection system comprising a plurality of coulometrically efficient electrochemical cells, in series, for sequentially oxidizing and reducing selected substances in a sample solution under controlled conditions prior to measurement on a downstream testing electrode or electrodes. More specifically, in accordance with the invention provided in my aforesaid U.S. Pat. No. 4,511,659, a sample solution (e.g. a body fluid) is passed through a suitable chromatographic column and the eluant is streamed in contact with a series of electrochemically isolated, in-line coulometric electrodes operated under conditions so as to establish a series of “gates” for the sequential oxidation and reduction of substances in the sample solution whereby to screen (remove) selected interfering and electrochemically irreversible substances contained in the sample solution, while passing selected electrochemically reversible products for detection and measurement on a downstream electrode. The gate electrode series is follows in-line by one or more, preferably an array of six or more coulometric measuring electrodes, each formed of porous electrode base material such as fritted graphite, fritted carbon or other conductive fritted material, for detecting and measuring the electrochemically reversible compounds of interest (e.g. neurotransmitters).

As reported in my aforesaid U.S. Pat. No. 4,511,659, there are several beneficial effects of this approach to electrochemical analysis. Long-term drift in response is effectively eliminated by acquiring essentially 100% of the signal. The capability of analyzing essentially 100% of a material allows the assay of compounds of unknown purity by relating them to the basic principles of electrochemical reaction embodied in Faraday's Law. Poisoning of the electrode, a dominant problem with electrochemical sensors, is effectively eliminated by the use of a much larger relative surface area for reaction. And, finally, and most important to the eventual development of array and gate cells, a coulometric electrode by virtue of its essentially 100% efficiency allows sequential oxidation and/or reduction of compounds at successive-in-line detectors. The improved sensitivity of the detection system as discussed in my aforesaid U.S. Pat. No. 4,511,659, particularly where two or more active testing electrodes follow the screening electrodes has given the ability to do direct injections of serum filtrates and has also allowed the generation of reproducible patterns of compounds with catecholamine like electrochemical behavior of a large number of resolvable components. This provides the possibility of performing pattern recognition for the diagnosis or perhaps even predictive diagnosis, of various disorders or disease states.

In my copending application Ser. No. 797,615 and its parent U.S. Pat. No. 4,863,873, I describe a system for resolving and detecting hundreds of compounds in a single sample at femtogram levels whereby to provide a small molecule inventory or metabolic pathway pattern of an individual. As taught in my aforesaid U.S. Pat. No. 4,863,873, the small molecule inventory may be considered to reflect the underlying activity and distribution of the enzymatic pathways of an individual and hence reflect an operational measure of the genome determining those enzymes. The small molecule inventory of an individual may thus be used to determine the health state of the individual and/or to diagnose disease states. Correlation of the patterns from a plurality of individuals provides an understanding of the mechanisms of disorders or disease states or conditions and, in turn, provides a rational route to pharmacological development leading to treatment, cure or suppression of such disorders, disease states or conditions.

The present invention is an improvement in the invention described in my aforesaid U.S. Pat. No. 4,863,873. More particularly, in the practice of my invention as described in my U.S. Pat. No. 4,863,873, I have observed that the biochemical analysis profiles of “normal” or healthy individuals may vary quite widely, while the biochemical profile analysis data of individuals having disorders is far less chaotic. More particularly, I have observed that the frequency distribution of certain biochemical compounds or ratios of compounds in individuals suffering from a disorder are far less chaotic than “normal” or healthy individuals. This leads to a general protocol for diagnosing, categorizing or differentiating individuals based on comparisons of biochemical analytical data of small molecule inventory against data bases of known or previously diagnosed cases. By way of example the process of the present invention may advantageously be employed in the differentiation of neurological degenerative dementing or affective disorders such as Alzheimer's disease, Huntington's disease, Parkinson's disease, schizophrenia, or Progressive Supernuclear Palsy from each other and neurologically normal controls. Moreover, by suitable selection of variables, the process of the present invention also is applicable to classification of tumors, carcinomas, cardiovascular abnormalities and other disorders. Similarly, the process of the present invention advantageously may be utilized to select therapy based on categories of known successful vs. unsuccessful outcomes.

While not wishing to be bound by theory, the two fundamental hypotheses underlying the process of the present invention are:

1. The underlying genetic makeup or predis-position of an individual will reflect through the proteins, enzymes, and other factors it determines in patterns of small molecules. Individual components within these patterns will be affected by environmental effects such as diet, stress or chemical inset; however, the overall pattern of relationships will reflect the underlying operation of the genome or the interference of a particular disorder. Among the small molecules are the transmitters, cofactors and metabolites that regulate neuronal and endocrine functions and the interactions of somatic and central nervous system processes. Thus, the compounds such as purines, tyrosine and tryptophan derived neurotransmitters, peptides, pterin and vitamine cofactors are highly relevant to the effect or etiology of neurological disorders, cardiovascular disfunction and certain tumors or carcinomas.

2. The relationships of these biochemical patterns from a disorder are less chaotic or more regular than those from healthy controls. All of the biochemical systems of small molecules are interconnected and interrelated in a complex web of feedback and response. These interactions are highly nonlinear and thus, depending on subtle differences in initial conditions, the response of individual components in a biochemical pattern will be highly variable. The overall system will thus behave in a mathematically chaotic fashion. In a disorder, elements within the biochemical pattern are over or underregulated, thus reducing the degrees of freedom or overall variability. Consequently, the presence of a disorder implies more regulated or less chaotic variability of compounds or relationships among compounds in patterns from disordered individuals.

These two fundamental hypotheses provide an approach to diagnostic categorization of disorders using frequency distributions of compounds and relationships from large data bases (which may be of epidermiologically significant size).

DESCRIPTION OF THE DRAWINGS

For a fuller understanding of the nature and objects of the present invention, reference should be had to the following detailed description taken in combination with the accompanying drawings wherein:

FIGS. 1 a - 1 d are plots showing frequency measurement distributions;

FIG. 2 is a plot showing sequential calculations;

FIG. 3 is a plot showing probability analysis; and

FIG. 4 is a plot similar to FIG. 3 , and showing probability analysis corrected for error.

DETAILED DESCRIPTION OF THE INVENTION

Methodology for Sample Analysis and Data Base Creation

280 cerebrospinal fluid (CSF) samples from the categories Alzheimer's Disease—AD (61 samples), Parkinson's Disease—PD (60 samples), schizophrenia—SC (60 samples), Huntington's Disease—HD (20 samples), Supernuclear Palsy—PSP (13 samples) and neurologically normal controls—C (68 samples), were electrochemically analyzed in accordance with the teachings of my aforesaid U.S. Pat. No. 4,863,873. Samples from normal and diseased individuals were prepared and flowed through a chromatographic column, and detected in an electrochemical cell using an NCA Chemical Analyzer, Model No. CEAS available from ESA, Inc., Bedford, Mass. All samples were from 7th or 8th mL aliquots of nostril caudal gradients. Pools were created for each category utilizing small subaliquots of the samples, and pools of all samples were created for analytical quality control and evaluation of unknowns. Samples were run under a variant of a standard reverse phase gradient procedure (Table 1) in the repetitive sequence Control Standard, Pool, 7 Samples, Control Standard, Pool, . . . as set forth in Table I:

TABLE I
REAL TIME SETUP REVIEW
	TIME	DEVICE	FUNCTION	VALUE	TOTAL
1	0.00	FLOW	%B	5	1.20
2	0.20	CLEAN CELL	ON	960
3	0.50	CLEAN CELL	OFF
4	7.00	AUTO ZERO	ON
5	7.20	FLOW	%B	5	1.20
6	7.58	AUTO SAMPLER	INJECT
7	8.00	FILE	START
8	66.00	FLOW	%B	94	1.06
9	66.00	FLOW	%B	5	1.20
10	70.00	FILE	STOP
11	70.00	FLOW	%B	5	1.20
12	70.00	AUTO SAMPLER	STEP
13	70.00	END		2
14	70.00	METHOD	M147
REVIEW OF LIVE METHOD
		Autorange
Full Scale Current	P'stats	On
10 uA	100 uA	10 uA	10 uA	−40 mV	25 mV	90 mV	155 mV
1 uA	1 uA	10 uA	1 uA	220 mV	285 mV	350 mV	415 mV	Floor
1 uA	1 uA	1 uA	1 uA	480 mV	545 mV	610 mV	675 mV	100 nA
1 uA	1 uA	1 uA	1 uA	740 mV	805 mV	870 mV	910 mV
Cell Box Temp: 35° C.
		UPPER LIMIT	LOWER LIMIT
	PUMP A:	350	0
	PUMP B:	350	0
	VALVE: POS1

TABLE II
Sat. Nov. 18 17:15:22 1990
	Standard  std0001
Table: NOY8STD
Study: CSFRUX01
# Identified Compounds in Standard: 38
# Missing Compounds in Standard: 0
	File	Name
#	RT	Conc	1	2	3	4	5	6	7
1	std0001	ASC	1.717	1.700	1.725	1.725	1.750	1.733
	1.725	10000.00	3872279	9625125	1.27e+07	1.60e+07	1.50e+07	1.14e+07
2	std0001	CYS							2.267
	2.283	1000.00							129467
3	std0001	URIC			2.333	2.342	2.350	2.350
	2.350	0000.00			5672	1324197	3193191	1019128
4	std0001	XAM
	3.058	1000.00
5	std0001	HX
	3.125	100.00
6	std0001	VMA				4.883	4.867	4.875	4.875
	4.875	10.00				387	4959	16674	13957
7	std0001	GSH							4.875
	4.900	500.00							13957
8	std0001	NE	5.625	5.642	5.642	5.650	5.650
	5.642	10.00	334	18928	17463	1507	117
9	std0001	MHPG					6.075	6.075	6.075
	6.075	10.00					749	13983	5545
	File	Name
#	RT	Conc	8	9	10	11	12	13	14	15	16
1
2	std0001	CYS	2.267	2.258	2.267	2.275	2.283	2.292
	2.283	1000.00	251043	528682	1298734	1601456	2250767	1062584
4	std0001	XAM		3.042	3.033	3.058	3.067	3.067	3.108
	3.058	1000.00		47171	1393667	3304889	2039171	350667	110609
5	std0001	HX							3.108	3.117	3.125
	3.125	100.00							110609	154070	161138
6
7	std0001	GSH	4.875	4.875	4.875	4.892	4.900	4.988
	4.900	500.00	28075	51532	117237	133772	197987	85618
8
9	std0001	MHPG	6.075	6.083	6.092
	6.075	10.00	1828	1718	1044
	File	Name
#	RT	Conc	1	2	3	4	5	6	7
10	std0001	HGA	6.392	6.408
	6.392	10.00	16607	4976
11	std0001	G
	7.942	100.00
12	std0001	GR
	9.150	500.00
13	std0001	LD	9.325	9.283	9.283	9.292	9.275
	9.283	10.00	165	10336	14158	1156	562
14	std0001	MET
	9.650	500.00
15	std0001	AM				10.008	10.008	10.017	10.000
	10.008	100.00				33745	118573	17512	2128
16	std0001	EPI	10.167	10.142	10.150
	10.142	10.00	367	14152	9783
17	std001	EPI	A
		10.217	100.00
18	std0001	DOPAC	11.825	11.833	11.833	11.842	11.850	11.833
	11.833	10.00	247	7794	9749	1412	445	251
	File	Name
#	RT	Conc	8	9	10	11	12	13	14	15	16
10
11	std0001	G		7.942	7.942	7.950	7.958	7.958
	7.942	100.00		2395	122100	53287	4420	975
12	std0001	GR			9.142	9.133	9.125	9.142	9.150	9.175	9.192
	9.150	500.00			859	26558	194121	430593	1054244	189936	111915
13
14	std0001	MET			9.642	9.642	9.650	9.650	9.650	9.667	9.675
	9.650	500.00			986	6461	26013	33548	107832	86595	31459
15
16
17	std001	EPI	A							10.250	10.217
	10.217	100.00								26046	36349
18
	File	Name
#	RT	Conc	1	2	3	4	5	6	7
19	std0001	30HAN		13.225	13.200	13.217	13.242	13.250
	13.200	10.00		1194	29905	8215	2107	1401
20	std001	30HXY		13.225	13.200	13.217	13.242	13.250
	13.200	10.00		1194	29905	8215	2107	1401
21	std0001	4HPLA
	13.450	100.00
22	std0001	XMN					13.683	13.692	13.683
	13.692	10.00					1593	14780	3917
23	std0001	4HBAC
	13.992	200.00
24	std0001	TYR							15.250
	15.233	1000.00							232
25	std0001	5HTP			15.583	15.592	15.617
	15.592	10.00			6994	16297	463
26	std0001	DA		17.342	17.342	17.358	17.367	17.375
	17.342	10.00		17722	6301	305	144	8313964
27	std0001	SHIAA		17.850	17.892	17.892	17.908	17.925
	17.892	50.00		233	9330	119759	14604	549
	File	Name
#	RT	Conc	8	9	10	11	12	13	14	15	16
19
20
21	std0001	4HPLA		13.442	13.450	13.467	13.458
	13.450	100.00		13364	125434	43118	2233
22
23	std0001	4HBAC				13.983	13.992	13.992	14.008
	13.992	200.00				75738	207495	311864	170423
24	std0001	TYR.	15.242	15.225	15.233	15.250	15.250	15.292
	15.233	1000.00	3866	102584	1181673	776764	117510
25
26
27
	File	Name
#	RT	Conc	1	2	3	4	5	6	7
28	std0001	4HPAC							18.083
	18.100	200.00							80
29	std0001	KYA
	18.117	50.00
30	std0001	30MD				21.575	21.608	21.608	21.600
	31.608	10.00				58	2183	11535	2176
31	std0001	5HTOL		22.167	22.200	22.200	22.217
	22.200	10.00		47	1668	19401	1931
32	std00l	HVA				23.925	23.925	23.917	23.917
	23.917	200.00				225	13994	237705	163964
33	std0001	KYX
	24.258	100.00
34	std0001	TYRA
	24.492	80.00
35	std0001	5HT		30.742	30.783	30.792
	30.792	10.00		231	6955	9627
36	std0001	3HT				31.967	31.975	31.975	31.967
	31.975	100.00				143	11501	33026	2425
	File	Name
#	RT	Conc	8	9	10	11	12	13	14	15	16
28	std0001	4HPAC	18.117	18.092	18.100	18.125
	18.100	200.00	1143	37258	461257	287050
29	std0001	KYA						18.075	18.117	18.167	18.175
	18.117	50.00						59412	112086	54548	38306
30	std0001	30MD	21.592
			143
31
32	std0001	HVA	23.925	23.925	23.950
			47372	8852	2293
33	std0001	KYX				24.267	24.258	24.283	24.300	24.308	24.308
	24.258	100.00				101620	108998	55261	39928	21718	4609
34	std0001	TYRA		24.475	24.492	24.517
	24.492	80.00		2368	28935	5390
35
36	std0001	3MT	31.975
			171
	File	Name
#	RT	Conc	1	2	3	4	5	6	7
37	std0001	TPOL							41.500
	41.467	10.00							375
38	std0001	TRP					45.350	45.342	45.388
	45.292	700.00					615	8371	23358
	File	Name
#	RT	Conc	8	9	10	11	12	13	14	15	16
37	std0001	TPOL	41.475	41.458	41.467	41.492
	41.467	10.00	2006	7109	9905	4810
38	std0001	TRP	45.300	45.275	45.292	45.325	45.350	45.350
	45.292	700.00	248497	967515	2056846	1090677	187013	109610

TABLE III
Tue Dec 18 11:43:57 1990
	Sample Report	Page 1
Sample: POOL37
Standard:
Table: POOL19A
Study: CSF19A
# Compounds identified: 22
# Known Compounds Not Found: 0
# Unknown Peak Clusters: 745
Compounds Identified
Com-
pound	Conc	RT	RT Error	Height	Ratio Accuracy
*p01	104.413971	2.617	0.100	167971	$$ 7 / 6 0.999
XAN	973.750732	2.900	0.117	756026	$$ 10 / 11 0.861 ⁢ 12 / 11 0.959
*p02	101.869156	3.108	0.125	208822	$$ 15 / 16 0.666
p03	131.687439	5.733	0.275	2747
p04	106.228569	8.242	0.325	142085	$$ 13 / 14 0.977 ⁢ 15 / 14 0.891
TYR	979.260254	13.567	0.692	1375195	$$ 9 / 10 0.987 ⁢ 11 / 10 0.977
*P05	99.553802	14.633	0.583	52515	$$ 13 / 14 0.949
P09	89.990860	15.117	0.650	7632	$$ 9 / 10 0.850 ⁢ 11 / 10 0.923
P07	92.967072	15.275	0.600	55664	$$ 14 / 15 0.951
P08	96.659180	16.133	0.417	29579	$$ 10 / 11 0.976
*P10	96.603462	21.875	0.942	42516	$$ 9 / 10 0.949
HVA	195.139450	22.475	0.800	30508	$$ 5 / 6 0.987 ⁢ 7 / 6 0.921
P11	102.689194	24.892	0.717	16233	$$ 16 / 15 0.892
P12	91.103188	25.150	0.850	10652	$$ 9 / 10 0.951
*p18	98.121086	26.850	1.308	470
*p19	121.804512	27.192	1.142	324
P13	96.415916	28.158	0.883	53718	$$ 12 / 13 0.960
P14	97.315338	29.542	1.108	24489	$$ 13 / 14 0.969
P15	94.104630	29.642	1.100	10631	$$ 9 / 10 0.916
P16	88.826569	32.425	0.767	2937	$$ 10 / 11 0.761
P17	96.782722	37.992	1.133	22787	$$ 13 / 14 0.958
TRP	674.284607	42.150	1.583	277999	$$ 9 / 10 0.979 ⁢ 11 / 10 0.966

The analysis records were linked by a unique identifier to clinical data of clinical diagnosis, diagnostic criteria, age, pharmaceutical history, sex and race. Pools analyzed as samples against standards for known values were utilized to assess the precision of known compound values in the data base. Standards sequentially analyzed against identical standards were used as a measure of instrumental performance and pools sequentially analyzed against identical pools were utilized as a measure of the precision of unknown peaks.

Validation of the Data

Control standards analyzed against sequential control standards yielded precision values ranging from ±1%-±4% CV (coefficient of variation) with no outlying values. Pools analyzed as samples gave precision values ranging from ±2-±7% CV for compounds present at the 0.5 ng/mL level or greater and typically ±25-30% for compounds present at 2× the detection limit of 0.02-0.03 ng/mL (e.g. 5HT, EPI). Pools analyzed against sequential pools for unknowns gave values of ±3-±15% coefficient of variation. Typically, the coefficient of variation of the pools was 5-25 fold less than the coefficient of variation of analytes in a group of samples. Essentially, the contribution of assay variability to the results is minimal.

The data base, upon completion, contained 280 samples by 57 analytes (17,000 records). Of these, 163 were null either because no peaks were detected at the sensitivity limits of the assay, or because a signal detected did not meet the qualitative criteria for purity.

Regression Analysis

Linear regression analysis and stepwise regression analysis were used in a preliminary evaluation of the data. Both raw and mean corrected data was evaluated.

Regression comparison of the AD group (61) vs. controls (60) setting AD=1 and C=0 gave a categorical separation regression equation with an S (standard error of estimate) value=0.39 and p (the probability that the sample belongs in one group and not in another) value=0.0041 for 27 of the most significant known compound variables identified in stepwise regression (Table I). Inclusion of 7 of the most significant variables (labeled with asterisks in Table II) from the pool analysed unknown peak data base gave values of S=0.382 and p=0.0037. Assuming a clinical diagnostic error rate in the order of 10%, seven AD samples with regression calculated values (from −1.2 to 0.01) were removed from the calculation. The regression characteristics were then S=0.352 and p=0.0031.

Regression of the AD group with AD=1 vs. all others (219)=0 for the same variable group yielded an equation with S=0.481 and p=0.0013.

Observations: Although the AD group is separated from other groups with a high degree of probability, there is too high a degree of overlap for a simple linear regression algorithm to accurately categorize an individual sample.

Cluster Analysis Procedures

Cluster analysis procedures using nearest neighbor arid furthest neighbor approaches were applied to the data base. With both these approaches, the AD group tended to cluster, but controls were scattered relatively evenly, both outside and inside the AD region. Thus, the cluster analysis approach is not suitable as a categorization tool for this type of data.

Observations: The behavior of the data under cluster analysis protocols, and the observations that the standard deviations of compound values and of precursor/product ratios across metabolic pathways within a disorder group are smaller than within control groups is consistent with the hypothesis that the biochemical response of controls or normal individuals is more chaotic than that of disordered individuals.

Frequency Distribution Probability Analysis

The observations on the nature of the data distributions coupled with the technical ability to run large numbers of samples and variables offers an approach to categorization based on differences in the frequency distributions of variables in different disorder categories. This approach relies in basic probability considerations without any assumptions on the shape of a distribution curve of linearity of relationships.

The simplest question that I have investigated for the preliminary data base is that given an unknown sample, what is the probability (p) that that sample belongs in one group and not another.

For one variable, the question takes the form: $$ P = f ⁡ ( V 1 ) A f ⁡ ( V 1 ) A + f ⁡ ( V 1 ) B

where F(V n ) A or f(V n ) B =the frequency with which an unknown sample value (V n ) occurs in category A or category B.

For multiple compounds, the expression expands: $$ P = f ⁡ ( V 1 ) A · f ⁡ ( V 2 ) A ⁢   ⁢ … ⁢   ⁢ f ⁡ ( V n ) A f ⁡ ( V 1 ) A · f ⁡ ( V ) A ⁢   ⁢ … ⁢   ⁢ f ⁡ ( V n ) A + f ⁡ ( V 1 ) B · f ⁡ ( V 1 ) B ⁢   ⁢ … ⁢   ⁢ f ⁡ ( V n ) B ⁢  

If all frequencies are the same, the P value is 0.5 or a 50/50 chance that the unknown sample is A and not B. A positive answer compresses the expression to a 1 and a negative answer to 0.

Like cluster procedures and unlike regression, the use of the algorithm is independent of the number of variables used.

Implementation of the Procedure:

The implementation of the procedure is by the following steps:

1. Frequency distributions (shown in FIGS. 1 a - 1 d ) were created by using a smoothing algorithm based on a 3 point polynomial expansion function that treats each point in the sparse data distribution with equal weighting as the means of a distribution with a width at half height proportional to its value. The use of smoothing functions is a necessary assumption until the number of samples in a particular category reaches approximately 300-400. The procedure used was to divide all data in the categories by the maximum value among categories X 85, apply the polynomial expression algorithm, and normalize the data distributions for the number of samples in each category. The frequency distributions in each category are then organized into look up tables for each variable (Table IV)

TABLE IV
DISTRIBUTION TABLE
	HVA			MHPG
	AD	CONTROL		AD	CONTROL
1	0.010377	0.000205	1	1.2E-10	0.000000
2	0.045259	0.001702	2	0.000000	0.000003
3	0.126131	0.007945	3	0.000000	0.000026
4	0.242295	0.025793	4	0.000000	0.000140
5	0.379120	0.064431	5	0.000000	0.000576
6	0.503888	0.129976	6	0.000001	0.001932
38	0.798418	0.938551	38	1.117459	0.742461
39	0.761982	0.889793	39	1.160736	0.801194
40	0.730955	0.836392	40	1.196534	0.855713
83	0.006846	0.088021	83	0.225452	0.069938
84	0.005088	0.081885	84	0.208132	0.064130
85	0.003699	0.075893	85	0.190170	0.041177

2. A sample record such as that shown in Table 5

TABLE V
DATACASE RECORD
		C	DISTRIBUT
		AT0022	RANGE
	MT3	0.018080	1
	OHAN3	0.018080	3
	OHKY3	0.090402	4
	3OMD	1.970763	44
	HVA	58.03809	56
	MHPG	8.714754	62
	12	2911604	22
	P01	5725159	9
	P02	1359827	49
	P03	1808040	1

is then inserted into the look up table. Individual values are divided by the range value X 85 and the frequencies for each variable for category A and B are sequentially calculated in the algorithm after subtracting the effect of that sample from the frequency table. The effect of a sequential calculation across a group of variables is shown in FIG. 2 for 3 AD and 3 C cases from a group of 61 AD and 44 controls. One of the major features of the algorithm is that no single variable predominates as a differentiation among a large group of samples.

Testing the Algorithm on AD vs. Controls

For an initial test, conditions were set up such that each individual sample was evaluated as if the data base were set up without its contribution. The results of the initial scoring are shown in FIG. 3 . The scoring of five of the 61 AD cases as controls (p=less than 0.01 that the sample is an AD and not a control) is not surprising given the probable diagnostic error rate in AD. The scoring of 4 of the controls as AD are of concern.

One possible explanation is that the AD data base is in effect contaminated by five cases that clearly do not match the overall AD profile and are probably not AD. When these five samples are removed from the AD data base and all samples, including the 5 removed, are scored, the control and AD groups are uniquely separated as shown in FIG. 4 . The five samples that were removed from the AD scoring data group distribute in an equivocal region from 0.1 to 0.9. In subsequent application of the procedure and algorithm to AD samples vs. all other samples (PD, SC, HD, PSP and C) in the data base asking the question is this sample in the AD distribution and not in the distribution of all others yielded similar results scoring AD samples with p values=0.98 or greater. The distribution of scores of all others was scattered from 0.001 to 0.8 including the 5 AD samples which previously scored in this region vs. controls.

The invention has been described for use in diagnosing Alzheimer's Disease from CSF patient samples. It will be understood, however, that the invention advantageously may be used to diagnose and characterize other neurological, degenerative or defective disorders such as Huntington's Disease, Parkinson's Disease, schizophrenia, progressive supernuclear palsy, ALS amyotrophic lateral sclerosis (Lou Gehrig's Disease) and senile dementias. The invention also advantageously may be used to classify and diagnose tumors, carcinomas, cardiovascular abnormalities and other disorders, or for selection of therapy based on categories of known successful vs. unsuccessful outcomes. Moreover, both treatment protocols and new pharmaceuticals may be evaluated.

Still other changes and advantages will be obvious to one skilled in the art.

Claims

1. In a method for diagnosing disorders in a test individual in which fluid samples from normal, unafflicted control individuals, afflicted, abnormal individuals, and said test individual are analyzed to generate electrical signal patterns representative of molecular constituents of said samples, the improvement which comprises creating a data base of electrical signal patterns representative of frequency distribution of a plurality of predetermined molecular constituents of fluid samples from an epidemiologically significant number of individuals having known categories of disorders and from said unafflicted control individuals, and comparing said electrical signal patterns in said data base for conformity to electrical signal patterns representative of frequency distribution of said predetermined molecular constituents of a fluid sample from said test individual.

2. A method according to claim 1, wherein said fluid samples comprise a body fluid.

3. A method according to claim 2, wherein said body fluid comprises cerebrospinal fluid.

4. A method according to claim 2, wherein said body fluid comprises plasma.

5. A method according to claim 2, wherein said body fluid comprises blood containing platelets.

6. A method according to claim 2, wherein said body fluid comprises nasal mucosa.

7. A method according to claim 2, wherein said body fluid comprises serum.

8. A method according to claim 2, wherein said body fluid comprises saliva.

9. A method according to claim 2, wherein said body fluid comprises urine.

10. A method according to claim 1, wherein said fluid samples comprise electrochemically active compounds, and wherein each electrical signal pattern representative of frequency distribution of said plurality of predetermined constituents of said fluid samples is generated by the following steps, comprising: passing each one of said fluid samples separately through a liquid chromatographic column for achieving time-space separation of the electrochemically active compounds of said fluid sample eluting in the column and generating electrical signals representative of the electrochemical pattern of said fluid sample using an electrochemical detection apparatus.

11. A method according to claim 1, wherein one of said known categories of disorders comprises Alzheimer's Disease.

12. A method according to claim 1, wherein one of said known categories of disorders comprises Parkinson's Disease.

13. A method according to claim 1, wherein one of said known categories of disorders comprises Huntington's Disease.

14. A method according to claim 1, wherein one of said known categories of disorders comprises schizophrenia.

15. A method according to claim 1, wherein one of said known categories of disorders comprises Progressive Supernuclear Palsy.

16. A method according to claim 1, wherein one of said known categories of disorders comprises amyotrophic lateral sclerosis.

17. A method according to claim 1, wherein one of said known categories of disorders comprises senile dementis.

18. A method according to claim 1, wherein one of said known categories of disorders comprises tumors.

19. A method according to claim 1, wherein one of said known categories of disorders comprises carcinomas.

20. A method according to claim 1, wherein one of said known categories of disorders comprises cardiovascular abnormalities.