Original Article
Lack of Efficacy of Water-Suppressed Proton Nuclear Magnetic Resonance Spectroscopy of Plasma for the Detection of Malignant Tumors
N Engl J Med 1990; 322:953-958April 5, 1990
- Abstract
- Abstract
Water-suppressed proton nuclear magnetic resonance (NMR) spectroscopy of plasma has been proposed by Fossel et al. (N Engl J Med 1986; 315:1369–76) as a technique for detecting malignant tumors. In their analysis, plasma samples from patients with cancer were clearly distinguished from those of normal subjects by measuring and averaging the methyl and methylene line widths of plasma lipoproteins in NMR spectrums.
To evaluate this diagnostic procedure further, we collected and analyzed by NMR spectroscopy 145 samples of plasma from patients who served as controls, most of whom were undergoing orthopedic or cardiac surgery (n = 66); patients with a variety of untreated malignant tumors (n = 25) or treated malignant tumors (n = 18); and patients with hyperplastic or "premalignant" diseases, such as benign prostatic hyperplasia and ulcerative colitis (n = 36). All the samples were coded, and NMR spectroscopy was performed without knowledge of the patients' clinical status.
There were no significant differences in the NMR line widths among the four study groups (P>0.05 for all pair-wise comparisons). The specificity and sensitivity of this method for distinguishing the control patients (mean line width [±SD], 44.0±7.4 Hz) from those with untreated cancer (43.8±6.9 Hz) were poor, with a false positive rate of 52 percent (34 of 66) and a false negative rate of 56 percent (14 of 25). Inverse correlations of line widths with age (P<0.01) and with the plasma triglyceride level (P<0.001) were detected.
We conclude that NMR spectroscopy of plasma is not an accurate test for the detection of malignant tumors. (N Engl J Med 1990; 322:953–8.)
Article Activity
- Article
THE concept that water-suppressed proton nuclear magnetic resonance (NMR) spectroscopy of blood plasma could be used to detect the presence of malignant tumors has been controversial. Fossel et al.1 2 3 4 5 measured the average line widths of the methyl and methylene resonances of lipoproteins in human plasma. They reported that this simple measurement alone could distinguish patients with virtually any type of malignant condition from healthy controls and patients with most nonmalignant illnesses. Other groups have attempted to reproduce these findings, with limited and varying success.6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 Many experimental studies have shown a difference between the mean line widths found in plasma from patients with cancer and from control subjects7 , 11 , 13 14 15 16 , 18 19 20 21; however, except for those performed at Beth Israel Hospital in Boston,1 2 3 4 5 , 21 none duplicated the original study in terms of the specificity and sensitivity of the test.
Several variables influencing the 1H NMR line widths of plasma lipoproteins have been identified.1 , 2 , 17 , 20 , 22 23 24 On the basis of these findings, the techniques used in some reports can be criticized for not strictly adhering to the protocol established by Fossel et al.1 2 3 For example, the specimens may have been frozen,2 hemolyzed samples or inadequately anticoagulated plasma may have been used, the magnetic field strength may have been too low, the temperature of the specimens during NMR analysis may not have been correct, the homogeneity of the magnetic field may not have been such that the proton line width of water was less than 4 Hz before water suppression, high plasma triglyceride levels (which characterize certain regional populations) may have caused false positive results, and finally, some of the specimens from patients with untreated cancer may have been improperly classified or their spectrums could have returned to normal as a result of excisional biopsy before blood was drawn. We therefore undertook a study to reassess the diagnostic accuracy of NMR spectroscopy by carefully following the procedures described by Fossel et al.1 and investigating factors that affected the test results. In addition, we used a gaussian model to estimate the frequency of false positive and false negative results from line-width distributions published by other investigators.
Methods
Collection of Blood
Blood samples collected before surgery in anticoagulant VenoJect tubes containing 5 ml of liquid EDTA (Terumo Medical, Elkton, Md.) were obtained from the general hematology laboratory at the Massachusetts General Hospital. Blood was acquired after all requested tests had been completed. The plasma was separated within 24 hours and stored at 4°C in polypropylene test tubes. Specimens with visible hemolysis were excluded. The samples were not frozen at any time. The coded samples were transported from Boston to the National Center for Toxicological Research in Jefferson, Arkansas, by one of us. During transport, the specimens were adjacent to but not in contact with ice, and no freezing was possible. One hundred thirteen specimens (78 percent) were analyzed within 6 days of collection, and all analyses were completed within 12 days. The duration of storage was well within established guidelines.1 , 21 EDTA peaks were visible in the NMR spectrum for each sample, confirming the presence of sufficient anticoagulant.
Classification of Patients
The patients from whom the preoperative blood samples were drawn included patients with confirmed untreated cancer, those with previously treated cancer, those with confirmed hyperplastic or "premalignant" diseases, and control patients undergoing unrelated surgery, mainly orthopedic or cardiac. The patients were selected according to the operating-room schedule published the day before surgery and were subsequently classified in one of the four groups on the basis of their clinical histories. Most of the patients with cancer were scheduled for biopsy or definitive surgery. The diagnoses were confirmed by the surgical pathology reports, and previous treatment was determined from the patients' medical records. The group with hyperplastic or premalignant disease consisted of patients with benign prostatic hyperplasia, Hashimoto's thyroiditis, ulcerative colitis, lipoma, or a history of dysplasia or severe atypia on a pathological study but no diagnosis of invasive cancer. The control group consisted mainly of patients undergoing orthopedic or cardiac surgery. Pregnant women were excluded. The medical history included information on current medications, the complete blood count, any recent weight change, sex, and age. The medicines taken were divided into four categories: nonsteroidal antiinflammatory agents, steroids, nitrates, and other medications. The investigators who carried out the NMR analyses were blinded to the patients' clinical status, and the clinical investigators were blinded to the plasma line-width data. The code was not broken until all clinical data were obtained and all NMR analyses were completed.
500-MHz NMR Analysis
NMR measurements at 20°C were made in the proton configuration at 500 MHz on a Bruker AM500 NMR spectrometer using Bruker DISNMR software (Rheinstetten, Federal Republic of Germany). The samples consisted of 600 μl of plasma and 20 μl of deuterium oxide in NMR tubes (diameter, 5 mm). To minimize line broadening due to the radiation-damping effects of the intense water signal, the observation channel on the probe was detuned just enough to attain a minimal line width of the water peak in a conventional spectrum. A flip angle of 75 degrees was used (55 μsec). The decoupler power used for presaturation was set so that a selective 180-degree pulse was 23.5 msec. The exact irradiation frequency for the water resonance was determined for each sample individually, and the duration of presaturation was set to four seconds. The spinning rate was about 17 rotations per second. Fine shimming (optimal magnetic-field homogeneity) was carried out on the water free-induction decay before spectrums were obtained. The line width of the water peak was measured and used as a criterion to evaluate magnetic-field homogeneity. Data were acquired with a sweep width of 10 kHz. Two dummy scans were followed by 16 scans.
The NMR data were processed with exponential multiplication that produced line broadening of 2 Hz. Since the line-width measurements were sensitive to spectral phasing, the same phase correction was used for each sample. The resulting base lines were nearly flat with respect to the end points and slightly elevated at the lowest point between the methyl and methylene resonances. The relevant region of the spectrum (0.43 to 1.6 ppm) was plotted with a scale of 12.5 Hz per centimeter. Apparent methyl (0.85 ppm) and methylene (1.25 ppm) line widths at half the height were measured from plots of this spectral region; the base line was defined as in Fossel et al.,1 and the measurements included the 2-Hz broadening function. The line widths were measured at half the height by drawing a line parallel to the base line. The lactate resonances were excluded from the methylene line width.
Validation of the NMR Methods
All the technical details were in accordance with methods verified in a previous study.25 In that project, carried out in collaboration with Fossel and with use of plasma samples supplied by him, we obtained better separation between line-width distributions at 500 MHz than did Fossel et al.1 at 360 MHz. Measurements made at 500 MHz were therefore judged to be nearly optimal, and better than those obtained at 360 MHz, for separating clinical groups.2 , 22
360-MHz NMR Analysis
Since a 360-MHz NMR spectrometer was used in the original study,1 we made additional measurements at this field strength. This portion of the study was carried out with a smaller set of samples (n = 40) and with use of a NMR spectrometer located at the Massachusetts Institute of Technology. The acquisition of specimens was similar to that described above; however, many specimens were collected from untreated private patients of one of us or from normal subjects in our laboratory. As in the report by Fossel et al.,1 presaturation for one to six seconds was sufficient for water suppression, no deuterium oxide was added, and the analysis was carried out at 20°C. Eight to 16 free-induction decays were averaged. Spectral processing and line-width measurements were then performed as previously described.
Statistical Analysis
The line widths for each clinical group are reported as means ±SD. The differences among groups were calculated with the two-tailed t-test, and simple linear regression with analysis of variance was used to examine for correlations of line width with continuous variables.
Results
500-MHz Studies
The analysis of the data from each group as a whole indicated little distinction among the groups (Table 1Table 1
Proton Nuclear Magnetic Resonance Line Widths of Lipoproteins (Measured at 500 MHz) and Patients' Age, According to Clinical Group.*, Fig. 1Figure 1
Distribution of the Average Line Widths at 500 MHz in Plasma Samples from the Four Clinical Groups.). The average line width in the samples from the control group differed by only 0.2 Hz from that in the group with untreated cancer. This represents a difference on the order of 1/30 of 1 SD. The false positive and false negative rates depended on the choice of discrimination points. When a reasonable discrimination point was chosen midway between the means of the control and untreated-cancer groups, the data yielded a false positive rate of 52 percent (34 of 66) and a false negative rate of 56 percent (14 of 25). It has been found that elevated plasma triglyceride levels increase the frequency of false positive results.2 , 15 , 16 , 22 23 24 , 26 However, the exclusion of samples with triglyceride levels above 2.3 mmol per liter (200 mg per deciliter) did not substantially improve the diagnostic usefulness of the method, despite the tendency to exclude older control subjects (control group, 46.6±5.9 Hz [n = 48]; untreated-cancer group, 46.3±5.1 Hz [n = 20]; false positive rate, 44 percent [21 of 48]; false negative rate, 45 percent [9 of 20]). The use of the methyl and methylene line widths alone resulted in similarly poor discrimination between patients with untreated cancer and control patients (Table 1). The false positive rates were 39 percent (26 of 66) for the methyl line width alone and 53 percent (35 of 66) for the methylene line width; the false negative rates were 44 percent ( 11 of 25) and 60 percent (15 of 25), respectively.Table 2Table 2
Diagnoses and Surgical Procedures Undergone by Patients in the Untreated-Cancer and Control Groups.* shows the plasma line widths as a function of the type of disease. The number of cases in most categories of untreated cancer was small, and the data therefore could not be subjected to statistical scrutiny; there was no evidence, however, that any subgroup of untreated cancer had an abnormal line width. One patient with two untreated primary malignant tumors (adenocarcinoma of the prostate and anaplastic astro-cytoma) had an above-average line width of 44.9 Hz. The narrowest line width in the untreated cancer group (29.5 Hz) was found in plasma from a patient with a pancreatic adenocarcinoma. The two broadest line widths in this group were in samples from a 51-year-old patient with an anaplastic astrocytoma (55 Hz) and a 24-year-old patient with metastatic osteosarcoma (53.7 Hz).The majority of the patients in the control group were undergoing orthopedic (n = 33) or cardiac (n = 14) surgery. Seven patients underwent litho-tripsy or surgery for renal calculi. There was no significant difference between the patients in any of the individual disease categories and the control group as a whole. One 29-year-old patient with uterine fibroids and endometriosis had an unusually broad line width (60.5 Hz). The narrowest line width in the control group was 26.0 Hz, in a 43-year-old patient with hypercholesterolemia from whom a renal calculus was removed.
There was an inverse correlation of the average line width with age in the combined study groups (correlation coefficient, 0.25; P<0.01). The maximal average line width occurred in the patients who were less than 40 years of age (48.2±7.6Hz [n = 33]); the line width decreased in the patients who were 40 to 59 years of age (42.1±8.1 Hz [n = 35]) and remained low for the patients 60 years of age or older (42.3±6.5 Hz [n = 77]). Similar patterns of line-width variation with age were seen in each of the groups. No significant correlations of line width with sex, hematocrit value, weight loss, or use of steroids, nitrates, or nonsteroidal antiinflammatory agents were found in this set of patients. Patients with plasma triglyceride levels ≥2.3 mmol per liter (≥200 mg per deciliter), however, had significantly narrower line widths (35.6±4.8 Hz [n = 39]) than those with lower plasma triglyceride concentrations (46.5±6.1 Hz [n = 106]; P<0.001).
The line width of the water peak before water suppression and that of the EDTA peak (chemical shift, 3.6 ppm) in the NMR spectrum were measured in order to evaluate the homogeneity of the magnetic field. It has been reported that the proton line width of water must be 4 Hz or less (at 360 MHz) and that a well-shimmed sample would result in a line width of 2.2 Hz or less for EDTA (at 360 MHz).2 The mean and standard deviation of the water line width and the EDTA line width in our samples, as measured at 500 MHz, were 3.1±0.3 Hz and 2.2±0.3 Hz, respectively. With one exception (4.2 Hz), all the water line widths were less than or equal to 3.7 Hz, indicating that very good magnetic-field homogeneity was achieved in this study. No correlation was detected between the line width for water or EDTA and the lipid line width.
360-MHz Studies
The mean line widths of the samples from the control subjects (n = 29) and from those with untreated cancer (n = 11) were 36.1–4.0 Hz and 33.8–3.8 Hz, respectively. Here again, the difference between the line widths of the patients with cancer and those without apparent cancer was substantially less than 1 SD. There was a significant correlation between the mean line width and age (P<0.05), but no correlation between the mean line width and the presence of a malignant tumor. The false positive rate was 31 percent (9 of 29) and the false negative rate was 36 percent (4 of 11 ) when we used a discrimination point of 34.9 Hz.
Frequency of False Positive and False Negative Results
The rates of false positive and false negative results were calculated from the means and standard deviations of the plasma line widths reported in the literature. It was assumed that the line-width distributions were gaussian and the optimal discrimination point was midway between the two groups. The assumed discrimination point was reasonable for most studies, since the standard deviations for the control and untreated-cancer groups were usually similar. This model predicted a false positive rate of 49 percent and a false negative rate of 49 percent in the data we obtained at 500 MHz (Table 3Table 3
Estimated Frequency of False Positive and False Negative Results in Published Studies.*) — a prediction that was in fairly close agreement with the actual values (52 percent and 56 percent).We used the same model to estimate the false positive and false negative rates from available data in the literature (Table 3). Many published studies contain sufficient data for the calculation of estimates of the actual false positive and false negative rates (Table 4Table 4
Actual Frequency of False Positive and False Negative Results in Published Studies.*). These data are usually presented in histogram form; hence, sample values that fall in the same class width as the discrimination point are considered to be half above and half below this value.With the exception of the studies in which plasma samples were analyzed at Beth Israel Hospital,1 2 3 4 5 , 21 the estimated rates of misclassification were higher than in the original report,1 even when studies performed at 300 MHz or less were excluded. The assumption that the line-width distributions were gaussian was, on average, valid. The gaussian model did not consistently overestimate or underestimate the actual false positive or false negative rates, as would be expected if the line-width distributions were not symmetric — as was suggested by Fossel et al.1
Discussion
Although we have attempted to follow all the procedures described by Fossel et al.,1 2 3 we could not distinguish accurately between samples from patients with untreated cancer and those from other patients solely on the basis of the average NMR line width of lipoprotein lipids in plasma. The correlation of line width with age that we did observe can partially explain the differences in the rates of false positive and false negative results among laboratories. Thus, some studies may report initially promising results because of the older age of patients with cancer, as compared with laboratory personnel and other apparently healthy persons who commonly make up the control groups in these studies. In our study, the controls and the patients with untreated cancer were of similar ages, and there was no significant disease-dependent difference in line-width distributions between these groups.
A factor that can alter the line widths is the lipoprotein composition of plasma.17 , 20 , 24 An interrelation between high-density, low-density, and very-low-density lipoprotein concentrations and average line width has also been observed.22 Given the well-known association between the plasma triglyceride concentration and age, it is likely that the NMR test measures plasma lipid composition and is therefore only indirectly correlated with the patient's age.
In a detection assay, individual measurements can be classified in clinical groups only if the overlap between the distributions of these groups is small. This criterion was not met by the data collected in this study. The poor ability of this test to distinguish patients with untreated cancer from control patients is the overwhelming observation of all investigations in which the plasma analyses were not done by Dr. Fossel's laboratory (Table 3).
One report has suggested that discrimination between groups is improved when only the methylene line width is used.15 Our study failed to show this association with methylene, but it confirmed the other findings of Berger et al.15 that this individual resonance, like the averaged resonances, was not sufficient to differentiate between clinical groups.
In summary, we could not reliably detect the presence of malignant tumors when we screened plasma samples from subjects chosen from a hospital patient population known to have a high incidence of cancer. We do not believe that this test is useful for screening even high-risk populations for cancer. We cannot account for the discrepancy between the results reported by Fossel and his coworkers1 2 3 4 5 and our own.
Supported by grants (CA48096, CA13311, and RR00995) from the National Institutes of Health and by an American Cancer Society Career Development Award.
We are indebted to Dr. Steven Westgate for his assistance.
Source Information
From the Department of Radiation Medicine, Massachusetts General Hospital and Harvard Medical School, Boston (P.O., A.Z., J.K.); the Francis Bitter National Magnet Laboratory, Massachusetts Institute of Technology, Cambridge, Mass. (S.S., L.J.N.); and the Division of Biochemical Toxicology, National Center for Toxicological Research, Food and Drug Administration, Jefferson, Ark. (R.A.L., F.E.E.). Address reprint requests to Dr. Okunieff at the Department of Radiation Medicine, Massachusetts General Hospital, Fruit St., Boston, MA 02114.
- References
References
1
Fossel
Full Text | Web of Science | Medline2
Fossel
CrossRef | Medline3
Fossel
Full Text | Web of Science | Medline4
Idem. Discrimination of malignant tumors in breast and lung tissues by water-suppressed proton NMR spectroscopy of plasma. In: Proceedings of the Sixth Annual Meeting of the Society of Magnetic Resonance in Medicine, New York, August 17–21, 1987. Vol. 1. Berkeley, Calif.: Society of Magnetic Resonance in Medicine, 1987:40. abstract.
5
Idem. Use of water-suppressed proton NMR spectroscopy of plasma to detect breast cancer in patients with palpable and unpalpable (asymptomatic) lesions. In: Proceedings of the Seventh Annual Meeting of the Society of Magnetic Resonance in Medicine, San Francisco, August 20–26, 1988. Vol. 1. Berkeley, Calif.: Society of Magnetic Resonance in Medicine, 1988:367. abstract.
6
Dowd
CrossRef | Web of Science | Medline7
Parl
Full Text | Web of Science | Medline8
Ross
Web of Science9
Regan
Web of Science10
Nabholtz
CrossRef | Web of Science | Medline11
Buchthal
CrossRef | Web of Science | Medline12
Otvos
CrossRef | Web of Science | Medline13
Peeling
CrossRef | Web of Science | Medline14
Bjoro K, Jellum E, Grace D. The detection of malignant tumours by 1H NMR of plasma: possible interference of the results in the case of hypertriglyceridemia. In: Proceedings of the Sixth Annual Meeting of the Society of Magnetic Resonance in Medicine, New York, August 17–21, 1987. Berkeley, Calif.: Society of Magnetic Resonance in Medicine, 1987:60. abstract.
15
Berger
CrossRef | Medline16
Wilding
Web of Science | Medline17
Bell
CrossRef | Medline18
Otvos JD, Hayes LW, Jeyarajah E, Ma X, Janjan N, Freedman DS. Relationship between H-1 NMR analysis of blood plasma. In: Proceedings of the Seventh Annual Meeting of the Society of Magnetic Resonance in Medicine, San Francisco, August 20–26, 1988. Berkeley, Calif.: Society of Magnetic Resonance in Medicine, 1988:169. abstract.
19
Wright
Web of Science20
Chmurny
CrossRef | Medline21
Hofeler H, Scheulen ME, Schmidt CG, Fossel ET. Monitoring of nonseminatous testicular cancer by water-suppressed proton nuclear magnetic resonance spectroscopy of plasma. In: Proceedings of the Seventh Annual Meeting of the Society of Magnetic Resonance in Medicine, San Francisco, August 20–26, 1988. Vol. 1. Berkeley, Calif.: Society of Magnetic Resonance in Medicine, 1988:358. abstract.
22
Evans
Web of Science23
Evans 24
Holmes
CrossRef | Medline25
Evans FE, Levine RA, Fossel ET. Effect of field strength on average lipid line widths determined by water-suppressed proton NMR spectroscopy of plasma. In: Proceedings of the Seventh Annual Meeting of the Society of Magnetic Resonance in Medicine, San Francisco, August 20–26, 1988. Vol. 1. Berkeley, Calif.: Society of Magnetic Resonance in Medicine, 1988:366. abstract.
26
Mims
Full Text | Web of Science | Medline
- Citing Articles (16)
Citing Articles
1
E. P. Diamandis. (2010) Cancer Biomarkers: Can We Turn Recent Failures into Success?. JNCI Journal of the National Cancer Institute 102:19, 1462-1467
CrossRef2
Hector C. Keun. 2010. Metabolic Profiling for Biomarker Discovery. , 47-74.
CrossRef3
Joel D. Morrisett, James A. Hamilton. 2007. Lipoproteins. .
CrossRef4
Tracy L. Whitehead, Behjatolah Monzavi-Karbassi, Thomas Kieber-Emmons. (2005) 1H-NMR metabonomics analysis of sera differentiates between mammary tumor-bearing mice and healthy controls. Metabolomics 1:3, 269-278
CrossRef5
Ø. Fluge, K.S. Gilje, E. Sletten, O.M. Kvalheim, E. Skaarland, J.F. Halvorsen, M. Farstad, O. Søreide. (1996) Proton nuclear magnetic resonance spectroscopy of serum from patients with colorectal neoplasia. European Journal of Surgical Oncology (EJSO) 22:1, 78-83
CrossRef6
Mika Ala-Korpela. (1995) 1H NMR spectroscopy of human blood plasma. Progress in Nuclear Magnetic Resonance Spectroscopy 27:5-6, 475-554
CrossRef7
P.J.D. Foxall, M. Spraul, R.D. Farrant, L.C. Lindon, G.H. Neild, J.K. Nicholson. (1993) 750 MHz 1H-NMR spectroscopy of human blood plasma. Journal of Pharmaceutical and Biomedical Analysis 11:4-5, 267-276
CrossRef8
Sigmund E. Lasker, Michael J. Iatropoulos, Stephen S. Hecht, Bijaya Misra, Shantu Amin, Edith Zang, Gary M. Williams. (1992) N-ethyl-N-nitrosourea induced brain tumors in rats monitored by nuclear magnetic resonance imaging, plasma proton nuclear magnetic resonance spectroscopy and microscopy. Cancer Letters 67:2-3, 125-131
CrossRef9
Y. Hiltunen, M. Ala-Korpela, J. Jokisaari, S. Eskelinen, K. Kiviniitty. (1992) Proton nuclear magnetic resonance lineshape studies on human blood plasma lipids from newborn infants, healthy adults, and adults with tumors. Magnetic Resonance in Medicine 26:1, 89-99
CrossRef10
T. Engan, K. S. Bjerve, A. L. Høe, J. Krane. (1992) Proton magnetic resonance spectroscopy of fractionated plasma lipoproteins and reconstituted plasma from healthy subjects and patients with cancer. Scandinavian Journal of Clinical & Laboratory Investigation 52:5, 393-408
CrossRef11
Paul A. Bottomley. (1992) The trouble with spectroscopy papers. Journal of Magnetic Resonance Imaging 2:1, 1-8
CrossRef12
Y. Hiltunen, M. Ala-Korpela, J. Jokisaari, S. Eskelinen, K. Kiviniitty, M. Savolainen, Y. A. Kesäniemi. (1991) A lineshape fitting model for1H NMR spectra of human blood plasma. Magnetic Resonance in Medicine 21:2, 222-232
CrossRef13
Terje Engan, Jostein Krane, Stener Kvinnsland. (1991) Proton nuclear magnetic resonance spectroscopy measurements of methylene and methyl line widths in plasma: Significant variations with extent of breast cancer, duration of pregnancy and aging. NMR in Biomedicine 4:3, 142-149
CrossRef14
E. T. Fossel, F. M. Hall, J. McDonagh. (1991) C-13 NMR spectroscopy of plasma reduces interference of hypertriglyceridemia in the H-1 NMR detection of malignancy. Application in patients with breast lesions. Breast Cancer Research and Treatment 18:2, 99-110
CrossRef15
(1990) Nuclear Magnetic Resonance Spectroscopy of Plasma for the Detection of Malignant Tumors. New England Journal of Medicine 323:10, 677-681
Full Text16
Shulman, Robert, . (1990) NMR — Another Cancer-Test Disappointment. New England Journal of Medicine 322:14, 1002-1003
Full Text
