Document:AIDS Case Fatality Rates

From AIDS Wiki

NOTWITHSTANDING ANY OTHER NOTICE ON THIS PAGE, the material on this page is NOT available under the GNU Free Documentation License; in accordance with Title 17 U.S.C. section 107, it is posted in the manner of bulletin boards in schools and workplaces, to encourage public education and citizen awareness, without profit or payment, for persons and entities engaging in non-profit research and educational activities and purposes only.

25 May 2007
[postscript 19 July 2007]

It is often claimed that the best proof that HIV causes AIDS is because "the drugs are working" (Avert 2007a, Köhnlein 2001, Szalavitz 2000). Usually, it is meant that the so-called "HAART [Highly Active AntiRetroviral Therapy]" has led to a dramatic reduction in AIDS deaths, and because these drugs were "rationally designed" to inhibit HIV replication, this is ample proof that HIV causes AIDS.

Is the reduction in deaths truly due to the drugs? The argument frequently employed is to look at absolute numbers of AIDS deaths from year to year. This is the wrong number to look at. The relevant number is the one-year AIDS case fatality rate, i.e. the number of AIDS deaths in a given year in proportion to the prevalence of AIDS cases: "Case-fatality is a measure of the severity of a disease. It can also be used to measure any benefits of a new therapy: As therapy improves, case-fatality would be expected to decline." (Gordis 2000) This number may be interpreted as the "probability that a given AIDS patient will die in a given year". (See the notes on definitions and computations below.)

When this number is examined, the surprising fact is that the one-year AIDS case fatality rate in the United States peaked in 1984-1986 and has been declining since 1987! This decline in the "AIDS epidemic" predates not only the advent of HAART, but also the advent of AZT monotherapy! Thus, it is misleading to claim that a decline in the one-year AIDS case fatality rate – which itself predates all ARV therapy – can be attributed to ARV's, and it is even more misleading to claim that such a decline supports the HIV hypothesis.

The thoughtful reader may study the data below and then ask herself how the "AIDS epidemic" in the United States might have proceeded if:

1. AIDS patients had been treated for their individual diseases since the mid-1980s, and not exposed to the fatalistic HIV = AIDS = Death message (Lauritsen 1988);
2. AZT monotherapy (now universally regarded as a failure which killed many patients (Farber 2000)) had not been implemented beginning in 1987;
3. The CDC had not expanded its definition of AIDS in both 1985 and 1987, and again in 1993 (Koliadin 1996).

[In the table below, C = AIDS Cases, D = AIDS Deaths, CC = Cumulative AIDS Cases, CD = Cumulative AIDS Deaths, AP = AIDS Prevalence = CC − CD, CFR = One-Year AIDS Case Fatality Rate = (D/AP), O = Odds = (1 − CFR)/CFR. Please see the notes on definitions and computations below. All data come from Avert.org (Avert 2007b)].

Year	C	D	CC	CD	AP	CFR	O
Before 1981	100	30
1981	339	130	270	95	175	74.5%	0.34
1982	1,201	466	1,040	393	647	72.1%	0.39
1983	3,153	1,511	3,217	1,382	1,835	82.3%	0.21
1984	6,368	3,526	7,977	3,900	4,077	86.5%	0.16
1985	12,044	6,996	17,183	9,161	8,022	87.2%	0.15
1986	19,404	12,183	32,907	18,751	14,157	86.1%	0.16
1987	29,105	16,488	57,162	33,086	24,076	68.5%	0.46
1988	36,126	21,244	89,777	51,952	37,825	56.2%	0.78
1989	43,499	28,054	129,590	76,601	52,989	52.9%	0.89
1990	49,546	31,836	176,112	106,546	69,566	45.8%	1.19
1991	60,573	37,106	231,172	141,017	90,155	41.2%	1.43
1992	79,657	41,849	301,287	180,495	120,792	34.7%	1.89
1993	79,879	45,733	381,055	224,286	156,769	29.2%	2.43
1994	73,086	50,657	457,537	272,481	185,057	27.4%	2.65
1995	69,984	51,414	529,072	323,516	205,556	25.0%	3.00
1996	61,124	38,074	594,626	368,260	226,366	16.8%	4.95
1997	49,379	21,846	649,878	398,220	251,658	8.7%	10.52
1998	43,225	19,005	696,180	418,646	277,534	6.9%	13.60
1999	41,356	18,491	738,470	437,394	301,077	6.1%	15.28
2000	39,513	17,139	778,905	455,209	323,696	5.3%	17.89
2001	39,262	17,726	818,292	472,641	345,651	5.1%	18.50
2002	39,620	17,318	857,733	490,163	367,570	4.7%	20.22
2003	40,902	18,020	897,994	507,832	390,162	4.6%	20.65
2004	40,907	18,099	938,899	525,892	413,007	4.4%	21.82
2005	45,669	17,011	982,187	543,991	438,196	3.9%	24.76

Notes on definitions and computations

As has been my experience, definitions and terminology in the field of so-called "epidemiology" are often ambiguous and poorly-defined as compared to those in mathematics, statistics, actuarial sciences, professional survey research, and other exact quantitative sciences. For example, in the text cited above (Gordis), a "definition" of case-fatality rate is given: "Case-fatality rate (percent) = (No. of individuals dying during a specified period of time after disease onset or diagnosis) / (No. of individuals with the specified disease) × 100". There are a number of problems with this "definition".

For example, another conflicting "definition" of case-fatality rate is given on a different page: "[Case-fatality rate] is defined as the number of people who die of a disease divided by the number of people who have the disease. Given that a person has the disease, what is the likelihood that he or she will die of the disease? This differs from a mortality rate, in which the denominator includes anyone at risk of dying of the disease – both persons who have the disease and persons who do not (yet) have the disease, but in whom it could develop."

From this defintion, you might conclude that the only difference between mortality rate and case-fatality rate is the denominator. You would be wrong. Gordis defines annual mortality rate from lung cancer as follows: "Annual mortality rate from lung cancer (per 1,000 population) = (No. of deaths from lung cancer in one year) / (No. of persons in the population at midyear) × 1,000." Here, both the numerator and the denominator are different. In the "definition" of case-fatality rate, the numerator was "the number of individuals who died during a specified period of time after disease onset or diagnosis", i.e. only those individuals who received a diagnosis say, within a year previously, are counted. However, in the definition of mortality rate, all individuals who die of the disease are counted, regardless of when they were diagnosed. So, which is it? Is the original definition of case-fatality correct, or is the second definition correct? They can't both be correct.

The denominator poses problems as well. What does "number of individuals with the specified disease" mean? Does it mean prevalence, i.e. the total number of individuals with the specified disease? Or does it mean incidence, i.e. the number of newly-diagnosed individuals? Literally, the denominator appears to indicate prevalence. However, in almost every epidemiological document I have examined which purports to measure "case-fatality rates", it is the incidence definition which is used, not prevalence. In other words, the denominator is taken to be the number of individuals diagnosed in a specified period of time, not the "number of individuals with the specified disease". Thus, we witness a fundamental conflict between a basic definition as presented in a standard university textbook, and the use of such a definition in practice.

However, the problems are considerably worse than this. There is the ambiguous element of time. Note that, even to make a comparison between mortality rate and case-fatality rate, as Gordis has done above, one must have an element of time in case-fatality, and on top of that, an element of time that agrees with the mortality rate. For example, you might compare annual mortality rate with one-year case-fatality rate. It would make no sense to compare the two if case-fatality were independent of expressed time. Yet Gordis makes the astonishing admission: "The case-fatality rate does not include any explicit statement of time. [Note that this directly contradicts the "definition" originally given by Gordis above.] However, time is expressed implicitly, because case-fatality is generally used for acute diseases in which death, if it occurs, occurs relatively soon after diagnosis. Thus, if the usual natural history of the disease is known, the use of case-fatality refers to the period after diagnosis during which death might be expected to occur." I have no idea what it means to say that "time is expressed implicitly". "Implicit" expressions or formulations have absolutely no place when presenting data transparently for others to examine.

Such considerations are not academic. Up until roughly 1997 or so, the CDC presented case-fatality rate tables for AIDS. (After 1997, they stopped presenting this data. I leave it to the reader to guess why.) However, the case-fatality rates were presented cumulatively, which not only bypasses, but entirely confounds the question of time altogether: "Case-fatality rates are calculated for each half-year by date of diagnosis. Each 6-month case-fatality rate is the number of deaths ever reported among cases diagnosed in that period (regardless of the year of death [my emphasis]), divided by the number of total cases diagnosed in that period, multiplied by 100. For example, during the interval January through June 1982, AIDS was diagnosed in 415 adults/adolescents. Through December 1993, 389 of these 415 were reported as dead. Therefore, the case fatality rate is 93.7 (389 divided by 415, multiplied by 100)."

The problems entailed by such an approach should be evident. In the 1994 CDC surveillance report quoted above, the case-fatality rate from 1982 counts deaths which have occurred up to 11 years after diagnosis, while the case-fatality rate from 1990 only counts deaths which have occurred up to 3 years after diagnosis. For a "disease" like "HIV/AIDS", in which many individuals are known to survive years after diagnosis, such presentation renders any comparison between the numbers absurd. (Note that this fact, which is even more true today after someone can become an "AIDS patient" with a single CD4 count < 200 cells/mm³, goes against the standard model, reproduced on many government websites and in countless textbooks, that death follows diagnosis by a couple years; click here for a typical example.)

If the case-fatality rate really represents, according to Gordis, the "likelihood that he or she [a person with the disease] will die of the disease", then it is absolutely mandatory that the time periods used to compute different numbers be the same. To do otherwise would be like trying to compare two pairs of dice, by comparing the numbers of snake eyes rolled, by rolling the first pair 1,000 times and the second pair 10,000 times.

Another issue emerges. Gordis claims that "[case-fatality rates] can also be used to measure any benefits of a new therapy: As therapy improves, case-fatality would be expected to decline." If this is the case, it would seem strange only to consider incidence, which is how I have seen case-fatality applied in practice. Certainly, the entire population diagnosed, not just those newly-diagnosed, is eligible to receive the potential benefits of such a new therapy? Especially with a "disease" like "HIV/AIDS", where survival after "diagnosis" (even if "untreated") is not unusual?

This is why I have taken the approach which I have above: namely, to measure what I would consider (mathematically) to be the best interpretation of "the probability that a given AIDS patient will die in a given year", that is, the number of AIDS deaths in a given year divided by AIDS prevalence. Of course, it would be ideal to have access to complete information – the exact years of diagnosis and death (if it has occurred) for all AIDS cases since 1981. (Such information would only require roughly 25²/2 ≈ 300 numbers. Even to give such information by half-year would only require roughly 50²/2 ≈ 1,200 numbers.) Certainly the CDC has such data – when an AIDS patient dies, I find it hard to believe they have no record of the year (or half-year) of AIDS diagnosis – indeed, knowledge of such data is implicit in the formulation of much of the data the CDC published up until about 1997. Such lack of disclosure of complete information is nothing new at the CDC, though – certainly in 1985, the CDC had knowledge of every patient's "characteristics" (presumed transmission categories together with gender), but rather than choosing to report the numbers in each of the 2⁶ = 64 possibile combinations of characteristics, they chose a "hierarchical presentation" which hid relevant information (Lauritsen 1985).

Due to the fact, especially in the early years, that time from AIDS diagnosis to death was frequently less than a year, the cumulative figures CC and CD were computed using the mortality rate convention of using a midyear estimate: in the table above, this was computed by adding half the year's cases/deaths to all previous cases/deaths. The "epidemiologically" minded reader may object to this; while I have been able to find data on total deaths out of diagnosed cases within a given time period, and on total deaths occuring during a given year, I have not been able to find exactly what I am looking for: namely, the number of (still-living) patients previously diagnosed with AIDS at the beginning of year X, and of these, the number that died in year X. If anyone knows where to find this information, I'd love to know.

In summary, my faith in epidemiology was already rather slim after years of trying to understand the illogic, confusion, and ambiguity in CDC and WHO reports on HIV/AIDS. I thought that turning to a standard university textbook on epidemiology might help clear up some confusion. On the very first query I made (case-fatality rate), I found the confusion and ambiguity compounded, not cleared up.

Postscript: 19 July 2007: More Defective (Deceptive) Practices by the CDC

Recently, someone sent me an MMWR (Morbidity and Mortality Weekly) report published by the CDC which claims to offer support for the claim "the drugs are working" as I described above (CDC 2006):

This report highlights several major epidemiologic features of the U.S. HIV epidemic, including the decrease in overall AIDS incidence, the substantial increase in survival after AIDS diagnosis (especially since highly active antiretroviral therapy [HAART] became the standard of care in 1996), and the continued disparities among racial/ethnic minority populations.

However, this report does nothing to prove the claim that HAART has caused a "substantial increase in survival after AIDS diagnosis". Rather, the report confirms that the CDC has full knowledge of the years of HIV and AIDS diagnoses for patients, but is deliberately withholding this information and obscuring its presentation in misleading and deceptive analyses and graphs:

The analysis described in this report included 1) HIV/AIDS case reports (i.e., HIV infection with or without AIDS) from the 35 areas (33 states, Guam, and the U.S. Virgin Islands) with integrated, confidential, name-based HIV/AIDS surveillance of sufficient duration to produce reliable data (i.e., 2001-2004) and 2) AIDS case reports from the District of Columbia, the 50 states, and U.S. territories received by CDC through June 30, 2005. Cases of AIDS and HIV/AIDS were analyzed by year of earliest reported diagnosis of AIDS or HIV infection, respectively. [my emphasis]

To see the exact manner in which the CDC deceptively presents survival data in an attempt to support HAART therapy, let's take a close look at Figure 2 from this MMWR report:

Note that this figure presents (graphically) precisely the type of data that I plead for in numerical form above. It is a form of "case fatality" data, and the fact that the CDC was capable of producing such a figure is further proof that they do indeed have access to the "complete information" by year of diagnosis that I mention above. I could make a similar figure based on the data I derived above, except that my figure would be based on AIDS prevalence rather than incidence (year of diagnosis), I would include each year as a separate curve, and each curve would only project one year into the future.

Notice how the data are grouped, however. My analysis above suggests that "case fatality" peaked (or, equivalently, survival rate reached its lowest point) sometime around 1984-1986. Yet the CDC insists on grouping together the highly disparate years from 1981-1992 as a single data set. This is deceptive on at least four counts:

1. It obscures the fact that survival reached a low point for those AIDS cases diagnosed roughly around 1984-1986;
2. It disproportionately favors those diagnoses which occurred later in the time period, closer to 1992, because there were many, many more AIDS diagnoses in the early 1990s than in the early 1980s or mid-1980s;
3. It throws together four different AIDS "definitions" (pre-1982, 1982-1985, 1985-1987, and post-1987), creating analytic absurdity;
4. It throws together all pre-1993 AIDS cases into a single data set, obscuring the effect of the 1993 definition, which allowed those with CD4 count < 200 cells/mm³ but no clinical disease to be counted as "AIDS" cases: obviously, with so many healthy "AIDS" patients after 1993, the 1993 definition change by itself greatly increased survival rates.

I refuse to believe that epidemiologists and the producers of such reports at the CDC are so grossly incompetent not to be aware of such defects. On the contrary, such defects appear to have been deliberately manufactured to deceive the reader.

References

© 2007 by Darin Brown