The second part of the Public Health Service’s technical report of June 1955 is the part nobody quoted in the newspapers. It is forty pages of chemistry, probability and tables, and it is the closest thing there is to an account of why a vaccine that passed its tests could paralyse children. It names no firm. The five manufacturers whose vaccine had been released by the end of April appear in its tables as Laboratories A to E, and a sixth, which had not yet sold anything, as F.

The theory and the particles

Salk’s inactivation rested on the assumption that killing virus with formaldehyde is a first-order reaction: that in any interval the same fraction of what remains is destroyed, so that the logarithm of the live virus falls along a straight line, and the line can be extended with confidence into concentrations too small to measure. The report took the assumption seriously enough to say exactly what it promised. Starting from a million infective units per millilitre and reaching one unit per millilitre at three days, the line gave one unit per million millilitres at six days and one per million million at nine, “a concentration of living virus in the order of one tissue culture infective unit in each million tons of inactivated viral suspension at the end of nine days.”

Then it said why a virus suspension might not behave that way. A suspension is not a uniform solution; it is particles of different sizes and shapes, some clumped, some caught in the protein that formaldehyde itself coagulates and that the eye cannot see. At high concentration the reaction is dominated by the great bulk of particles that are easy to kill, and looks first-order. As those are used up, the resistant fraction asserts itself and the line bends. The report drew it: Curve A, a million easy particles per millilitre inactivating at the full rate; Curve B, a thousand per millilitre at half that rate; Curve C, 32 per millilitre at a quarter; and Curve D, their sum, which follows A for a while and then C. “Viewed in this light, it may be concluded that observations over the initial phase of the reaction cannot be taken in themselves as the basis for predicting the course of the remainder of the reaction.” Add to that uneven heating, formaldehyde poured in without proper mixing, and a vessel stirred less than it should be, all of which the inspectors found in one plant or another, and the margin of safety that was supposed to be built into the process could disappear without anyone having done anything that the requirements forbade.

The report also noted the choice of Type I strain. Mahoney had been selected for the vaccine because it provoked the most antibody. It was also, of the three strains, by far the most virulent, able to multiply in tissues outside the nervous system, so that a minute dose of living virus could produce a general infection. Denmark, making its own vaccine that spring, had excluded it for that reason and used the weaker Brunhilde strain with more antigen to compensate.

The two tests

Two tests stood between the inactivated pool and the bottle. The monkey test was applied to the trivalent pool only: twelve monkeys given a millilitre into each cerebral hemisphere and six given ten millilitres into the muscle, and watched for paralysis. Its sensitivity, the report said, was “crude, as was demonstrated in the spring of 1954,” and every manufacturer had found that it missed virus that tissue culture caught. Of seventeen trivalent pools found positive across the industry by May 1955, twelve had been caught by tissue culture alone, three by the monkeys alone, and two by both. The consultants’ recommendation was to make the monkey test more sensitive by concentrating the material or to stop relying on it.

The tissue-culture test was the better instrument, and it had never been asked what it could see. “It has not as yet been determined at what viral particulate concentration and under what circumstances a positive demonstration of live virus can be expected in a significant proportion of tests run.” The requirements of 12 April asked that a sample of at least one-tenth of one percent of the pool be tested: a millilitre for every litre. The report worked out, with the Poisson distribution and four stated assumptions that erred on the side of the test, what that meant. If a pool of ten litres or more held five infective units per litre, and a single particle in the sample were enough to show in a sensitive culture, then to keep the chance of a false negative down to one in ten one had to test 460 millilitres; to one in a hundred, 920; to one in a thousand, 1,380. If two particles were needed to show, the samples rose to 778, 1,328 and 1,846. At ten units per litre the sizes halved; at a hundred, they fell to tens of millilitres.

Against that the report set what each laboratory had actually been testing before 22 May. Laboratory A: 60 millilitres of each single-strain pool and 150 to 175 of each trivalent pool, which at five units per litre gave a 70 percent chance of catching a contaminated pool. Laboratory B: 500 and 360 millilitres, 99 percent. Laboratory C: 400 to 500 and 1,200 to 1,600, better than 99.9. Laboratory D: 25 and 300, 80 percent. Laboratory E: 140 for the trivalent pool, 60 to 70 percent. Laboratory F: 375 and 1,125, better than 99.9. “It is quite apparent that different standards of acceptability as regards the possible presence of infective particles have been inherent in the processes of the different manufacturers under the April 12 requirements.” The fault was in the rule, not only in the firms: a sample proportional to the pool is the wrong kind of sample. “Established statistical principles indicate, however, that to ensure with a high degree of probability that the concentration of live virus particles in any lot will be below a prescribed level, a sample of fixed minimum size must be tested, rather than one proportional to the size of the lot.”

What the records showed

The plant survey was made with the manufacturers’ full records, including the lots that had never been offered for release, and its tables are the first time anyone outside the firms had seen them. Across the five laboratories, 809 single-strain pools had been given the first tissue-culture test at the point where inactivation was supposed to be complete, and 11 percent had been positive. By laboratory: A, 97 pools, 21 percent positive; B, 234, 19 percent; C, 85, 8 percent; D, 267, 5 percent; E, 126, 2 percent. There was no difference between the three virus types. Laboratory C had been troubled by contamination of its cultures and some of its positives were probably not virus. “This is an important finding. It demonstrates that the departure of the process from one of systematic predictability greatly diminishes the factor of safety which was believed to be inherent in the basic process.”

Then the trivalent pools, each made from three single-strain pools that had already passed. Of 127 tested by tissue culture and monkeys, 17, or 13 percent, were positive: at A, 7 of 21, a third; at B, 3 of 34; at C, 5 of 22; at D, none of 8; at E, 2 of 42. “It will be seen that live virus is frequently found in a trivalent pool when this has been presumed to be free of live virus on the basis of testing the monovalent pools which compose it.”

What happened to a pool that failed is set out for Laboratory B. Of its 234 single-strain pools, 44 had been positive on the first test. They were put back at 37 degrees for another two or three days, after which 21 tested negative and were accepted. Of the rest, nine passed after a second reheating, two were still positive and were rejected, one was positive after three reheatings, seven were still in process, and four had been accepted on a repeat test of the same sample without any further treatment at all. Then the report followed those pools forward. Of 32 trivalent pools whose history was known, 23 had been made from single-strain pools negative from the start, and all 23 were negative. Nine had been made with at least one pool that had been positive and had passed after reheating, and three of the nine were positive again. “A single negative tissue culture test is poor evidence that reinactivation of a lot which gave a positive test after initial inactivation has caused the destruction of residual live virus.” Laboratory A’s record was similar: of 16 trivalent pools made wholly from pools that had passed first time, 3 were positive; of 10 made partly from reheated pools, 3. “In all, a quarter of all polyvalent lots, presumably inactivated, tested positive by tissue culture.” Laboratory D had reheated 13 positive pools and none had failed again, nor had any trivalent pool made from them. Laboratory E, which stirred its vessels vigorously and ran a highly uniform process, had 124 of 126 consecutive single-strain pools negative; but it tested small volumes, and one of its 41 trivalent pools made from clean components was positive anyway.

The report’s summary of all this is one paragraph. “In summary, the plant data show a departure from the theoretical in processing and a lack of discrimination in the current safety testing practices. There is inadequate assurance that vaccine so produced and tested will with any degree of certainty be consistently below some given level of virus content. There are exceptions to certain of these criticisms but these are not detailed since the examination of the data has been aimed primarily at discovering weaknesses in the process. Some of these are to be found in each of the laboratories.”

The Cutter lots, tested again

The report does not say which letter Cutter was, and the question of whether anything about the Berkeley plant was unique was left to the separate report of August, which this entry could not reach. What the record does show is what the released Cutter lots had looked like to the tests on the books. In the routine monkey tests run by Cutter and by the government before release, 74 monkeys were injected with vaccine from the lots later shown to carry live virus, and none was infected. After the cases appeared, eighteen laboratories tested the same lots by every route and with every trick that was known to make a monkey susceptible, cortisone and X-rays among them, and found type 1 virus in 46 of 461 monkeys, from seven of the eight filling lots drawn from the two bad production pools, and from none of the three good lots examined. Tissue culture, on 170 to 850 millilitres of each lot, gave one isolation. The virus was there, at a concentration the routine tests could not see, in two pools out of eight. Which is what the report’s Table 2 had predicted a 60-millilitre sample would do.

The new rules

The amended requirements of 25 May 1955 replaced the line with a number. A single-strain pool was to be inactivated in formaldehyde at one in four thousand, at 37 degrees, until it gave two consecutive negative tissue-culture tests on samples of at least 500 cubic centimetres each, three days apart; in practice that meant a first negative somewhere between the sixth and ninth day and a second three days later, and the earlier readings would show only that the process was proceeding regularly. A pool that tested positive could be reinactivated once, and once only, through the same two tests, “since additional periods of heating late in the process can be expected to lead to little in the way of further inactivation.” The trivalent pool had to pass on 1,500 cubic centimetres. A random sample from the final containers of every lot had to be tested, as a guard against anything that happened after the pool was cleared. The stated objective was a concentration of no more than five live units per litre, and the report’s estimate was that a vaccine passing all three stages would hold five or more units per litre “less than once in 100,000 times.” The makers were to report every lot they started. Under those rules, by the surveillance unit’s count, more than 400 million doses were given between 1955 and 1961 without a case of inoculation polio, and the epidemiologists’ later estimate for the first hundred million was that not more than one paralytic case had occurred per million inoculations.

One adviser did not sign. William Hammon of Pittsburgh agreed that the changes were advisable and was not satisfied that they gave an adequate margin of safety; he recommended that no more vaccine be released until production and laboratory methods and the epidemiology had been studied further. The report printed his dissent in a footnote and released the vaccine.

Sources

• United States Public Health Service, Technical Report on Salk Poliomyelitis Vaccine, June 1955, chapters 6 to 10 and their addendum tables, and chapter 5 for Hammon’s dissent. Read in the Global Health Chronicles scan of the National Archives copy. Global Health Chronicles
• Neal Nathanson and Alexander D. Langmuir, “The Cutter Incident,” II, American Journal of Hygiene 78, 1963, pp. 29 to 60, Table 16, for the tests on the Cutter lots and the 400 million doses. Global Health Chronicles
• Paul Meier, “Safety Testing of Poliomyelitis Vaccine,” Science 125, 1957, pp. 1067 to 1071, and Jerome Cornfield, Max Halperin and Felix Moore, “Some Statistical Aspects of Safety Testing the Salk Poliomyelitis Vaccine,” Public Health Reports 71, 1956, pp. 1045 to 1056: the later statistical treatments of the same problem, cited by Nathanson and Langmuir; not read.
• Report on Poliomyelitis Vaccine Produced by the Cutter Laboratories, Public Health Service, 25 August 1955; not reached.