The epidemic that arrived with a new codebook

Dark navy deck slide, masthead “Mortality Desk · United States · 1998 → 1999”, headlined “American death changed shape in a single year”. A subhead reads “Every death certificate filed in the country, both years, age-adjusted to one standard population. Total mortality barely moved — 875.8 per 100,000 to 881.9, up 0.7% — but underneath it the causes rearranged themselves. Alzheimer’s disease nearly doubled and climbed four places up the leading-cause table. Influenza and pneumonia lost a third of its toll in twelve months.” Under the label “Change in age-adjusted death rate, 1998 to 1999”, eight bars run from a zero line on an axis marked −25%, 0%, +25%, +50%, +75%, captioned “deaths per 100,000, year-2000 standard population”. Each row carries the two rates it is drawn from: Alzheimer’s disease 8.6 → 16.5, a long gold bar labelled +91.9%; kidney disease, noted as nephritis, nephrotic syndrome and nephrosis, 9.8 → 13.1, +33.7%; septicemia 8.9 → 11.3, +27.0%; hypertension 5.4 → 6.3, +16.7%; chronic lower respiratory diseases 42.0 → 45.8, +9.0%; stroke 59.5 → 61.8, +3.9%; all causes 875.8 → 881.9, a sliver of a bar at +0.7%; and influenza and pneumonia 34.6 → 23.6, a teal bar running the other way at −31.8%. Three ticked boxes along the bottom read “Every death, both years — 2,337,256 certificates in 1998 and 2,391,399 in 1999, nothing sampled, nothing modelled”, “Age-adjusted, one standard — the same year-2000 standard population on both sides of the comparison”, and “One office, one system — the same national vital statistics system published both years”. Down the right, a card reads “+91.9%, Alzheimer’s disease, age-adjusted death rate, 1998 to 1999. No other leading cause of death moved half as far.” A second card headed “The leading-cause table, 1998 → 1999” lists Alzheimer’s disease 12th → 8th in gold, diabetes 7th → 6th, septicemia 11th → 10th, hypertension 15th → 13th, influenza and pneumonia 6th → 7th and chronic liver disease 10th → 12th, noting that atherosclerosis left the fifteen leading causes after at least four decades and aortic aneurysm and dissection entered at fifteenth. A gold note headed “Nothing modelled here” says these are completed death certificates, not estimates — every death registered in the United States in 1998 and in 1999, age-adjusted by the same office to the same standard population; total mortality moved 0.7%, and what changed is what Americans died of. Beneath it, in small grey type: “Cause of death classified by the International Classification of Diseases; 1999 tabulations use the Tenth Revision.” A footnote credits the National Center for Health Statistics, “Deaths: Final Data for 1999”, National Vital Statistics Reports 49(8), 2001, table 8, and calls the slide a Mortality Desk demonstration. — Showing the misleading chart

The same two years on a cream background, kicked “The same two years · with the codebook change put back” and headlined “1999 was the first year America coded its deaths in ICD–10”. A subhead reads “Before the switch, NCHS coded some 1.85 million of the 2,314,690 deaths registered in 1996 twice — once under the Ninth Revision, once under the Tenth — and published a comparability ratio for every cause. Put those ratios on 1998 and most of the year’s drama turns out to be the codebook.” A legend marks blue as “change as published” and gold as “change with 1998 recoded to ICD–10”. Eight rows each carry the cause, its comparability ratio, and a pair of bars on one shared axis from −25% to +75%: Alzheimer’s disease, ratio 1.5536, +91.9% as published against +23.1% recoded; kidney disease 1.2320, +33.7% against +8.3%; septicemia 1.1949, +27.0% against +6.6%; hypertension 1.1192, +16.7% against +5.0%; chronic lower respiratory diseases 1.0478, +9.0% against +4.1%; stroke 1.0588, +3.9% against −1.9%; all causes, no ratio, +0.7% both ways; and influenza and pneumonia 0.6982, −31.8% against −2.5%. A note explains that the ratio is deaths coded to that cause under ICD–10 divided by deaths coded to it under ICD–9, on the same 1996 certificates. A panel headed “The total is the control” reads: a recode moves deaths between headings; it cannot change how many there were. All causes went 875.8 to 881.9 per 100,000, up 0.7%, and the ratios leave it there. When every heading jumps and the total does not, the headings are what changed. Three cards run down the right. “Where the pneumonia went” explains that ICD–10 widened the direct-sequel rule, so pneumonia written on a certificate alongside heart disease, stroke, cancer, chronic lower respiratory disease, septicemia, malnutrition or cirrhosis is now read as their consequence and the other condition is selected as the underlying cause; in the double-coded 1996 file influenza and pneumonia fell from 72,371 deaths to 50,526, a ratio of 0.6982, and the deaths that left pneumonia are most of what arrived at stroke, septicemia and chronic respiratory disease. “Alzheimer’s: a relaxed diagnosis rule” explains that under ICD–9 a definitive diagnosis was required to code a death to Alzheimer’s disease (331.0) and “Alzheimer’s-type dementia” went to presenile dementia (290.1) instead; ICD–10 dropped that stricture and about 95% of the gain came out of 290.1. In 1998 the United States recorded 22,725 deaths under 331.0 and 15,672 under 290.1 — add the two headings ICD–10 merged and 1998 holds 38,397, against 44,536 in 1999. “The conversion factor has its own error” notes that the ratios were measured on 1996 certificates while reporting of Alzheimer’s-type dementia kept growing, so NCHS says a ratio computed on 1998 data would be at least 1.69 and on 1999 data “could be as high as 1.8 or 1.9” — enough to erase the remaining 23.1% too; its own report calls that increase “may be spurious”, so the honest chart claims no 23% rise either, only that this pair of years cannot settle the question. A footnote credits NVSR 49(8) table 8 for the rates, notes that the 1998 column is ICD–9 data already tabulated on the nearest ICD–10 categories so what the ratios correct is the coding rules rather than the code list, credits Anderson, Miniño, Hoyert and Rosenberg’s comparability report for the ratios and the double-coded 1996 counts, and says the percentage changes are ours and both drawings our own. — Showing the misleading chart

Be the first to star this exhibit

Between 1998 and 1999 the American death rate from Alzheimer’s disease rose 91.9%, kidney disease 33.7% and septicemia 27.0%, while influenza and pneumonia fell 31.8% — and all causes together moved 0.7%. 1999 was the first year the United States coded its death certificates to the Tenth Revision of the ICD. NCHS had coded 1996’s certificates both ways and published the conversion factor for every cause.

National Center for Health Statistics, “Deaths: Final Data for 1999”, National Vital Statistics Reports 49(8), 2001, table 8 — with comparability ratios from Anderson, Miniño, Hoyert & Rosenberg, “Comparability of Cause of Death Between ICD–9 and ICD–10: Preliminary Estimates”, NVSR 49(2), 2001, 1999Added August 14, 2026

01The claim

American death changed shape in a single year. Every death certificate filed in the country is here, both years, age-adjusted to the same standard population by the same office — nothing sampled, nothing modelled, 2,337,256 certificates in 1998 and 2,391,399 in 1999. Total mortality barely moved: 875.8 per 100,000 to 881.9, up 0.7%. But underneath that flat total the causes rearranged themselves. The age-adjusted death rate from Alzheimer’s disease went from 8.6 to 16.5, up 91.9%, and the disease climbed from twelfth to eighth among American causes of death. Kidney disease rose 33.7%, septicemia 27.0%, hypertension 16.7%. Influenza and pneumonia lost a third of its toll, 34.6 down to 23.6. Atherosclerosis fell off the list of fifteen leading causes after at least four decades on it. Total mortality moved 0.7%; what changed is what Americans died of.

02The trick

Every figure on the slide is NCHS’s own, printed beside its bar, and the disclosure is on the slide too — in grey, under the last card: “1999 tabulations use the Tenth Revision.” 1999 was the first year the United States coded its death certificates to ICD–10 rather than ICD–9, and the two revisions do not just rename categories, they change the rules for choosing which condition on a certificate becomes the underlying cause. The largest of those, the direct-sequel rule, now treats pneumonia as a consequence of a much longer list of conditions, so a certificate listing pneumonia alongside heart disease, stroke, cancer, chronic lower respiratory disease, septicemia, malnutrition or cirrhosis is filed under the other condition instead. Alzheimer’s got a second change of its own: ICD–9 required a definitive diagnosis, and sent “Alzheimer’s-type dementia” to presenile dementia (290.1); ICD–10 relaxed that, and about 95% of Alzheimer’s gain came straight out of 290.1. None of this is guesswork — NCHS coded some 1.85 million of the 2,314,690 certificates registered in 1996 under both revisions and published a comparability ratio for each cause: 1.5536 for Alzheimer’s, 1.2320 for kidney disease, 1.1949 for septicemia, 1.1192 for hypertension, 1.0588 for stroke, 1.0478 for chronic lower respiratory disease, 0.6982 for influenza and pneumonia. The tell is the flat total the slide offers as reassurance. A recode moves deaths between headings; it cannot change how many there were. When seven headings jump by up to 92% and the total moves 0.7%, the headings are what changed. (This exhibit is our own demonstration, drawn in the house style of a mortality review from NCHS’s published tables — no such slide was published; NCHS printed the comparability-adjusted figures right alongside the raw ones.)

03The fix

Break the line where the definition broke, and where the publisher has measured the break, apply it. NCHS did the work: table 8 of the same report carries a “1998 modified” row for every leading cause, the 1998 rate recoded to ICD–10 through those ratios. Against that row the year looks entirely different. Alzheimer’s +91.9% becomes +23.1%. Kidney disease +33.7% becomes +8.3%, septicemia +27.0% becomes +6.6%, hypertension +16.7% becomes +5.0%, chronic lower respiratory disease +9.0% becomes +4.1%. Stroke’s +3.9% turns into a decline of 1.9%. Influenza and pneumonia’s collapse of 31.8% shrinks to 2.5%. All causes stays at +0.7%, because the total was never affected by any of it. Then hold the conversion factor to the same standard as the data, because it is a measurement and not a constant. NCHS says so itself about this one: the ratios were estimated on 1996 certificates, reporting of Alzheimer’s-type dementia kept climbing afterwards, and a ratio computed on 1998 data would be at least 1.69 while one computed on 1999 data “could be as high as 1.8 or 1.9” — enough to erase the remaining 23.1% as well. Its own report calls that increase “may be spurious”. So the honest chart does not claim a 23% rise in Alzheimer’s mortality either; it says this pair of years cannot settle the question, and points at the years on either side that can. The habit worth keeping is smaller than any of that: when a series steps, find out what changed that year before deciding what happened that year — and check the total, because the total is the control.