Comments to Ethical Skeptic (part 4)

Ethical Skeptic made this plot of cancer incidence in England, where he employed a hacky method to approximate an age-standardized incidence, so that he divided the number of cases in all ages with the population estimate of ages 65 and above: [https://x.com/EthicalSkeptic/status/1966911244859842912]

His plot has a massive drop in the incidence between the start of 2019 and the end of 2019, but in the dataset he cited as his source, the incidence remains roughly flat throughout 2019: [https://nhsd-ndrs.shinyapps.io/rcrd/]

I used WebPlotDigitizer to digitize the data in Ethical Skeptic's plot. Some points in his plot have irregular horizontal spacing, so I had to click on each point manually to extract the data correctly: [https://automeris.io/wpd/]

ES seems to have calculated the number of people aged 65 and above through a circuitious method, where he multiplied the yearly population estimates of England with the yearly proportion of people aged 65 and above:

I tried to replicate his methodology, so I took yearly population estimates from a page at Statista titled "Population of England from 1971 to 2023": https://www.statista.com/statistics/975956/population-of-england/. I used 2023 estimates for 2024 and 2025, but I'm not sure if for example ES used a linear projection of the 2022 and 2023 population sizes as the population size for 2024. I took the proportion of people aged 65 and above in England from the website of the ONS (where I used the 2024 proportion for 2025): https://www.ons.gov.uk/explore-local-statistics/indicators/percentage-of-the-population-aged-65-plus ("Line chart > Change areas > Other areas > All countries > England > Download line chart data (CSV)").

Then in order to calculate the yellow line in the next plot, I multiplied the monthly rates in Ethical Skeptic's plot with the yearly number of people aged 65 and above, and I divided the result by 100,000.

I downloaded the rapid cancer registration data from here: https://nhsd-ndrs.shinyapps.io/rcrd/. I used the version of the data adjusted for the number of working days per month, because the unadjusted version gave me a much poorer fit to Ethical Skeptic's plot, so he probably used the adjusted data. Then I took a backwards moving average of the data where the window extended up to 5 months backwards, but for example the window extended only 1 month backwards in February 2018.

I was able to replicate the values in Ethical Skeptic's closely in 2018. But he seems to have manually altered the y-axis values so that he shifted them downwards in 2019 to 2022, and he shifted them upwards in 2024 and 2025. His altered data has a particularly large downwards shift in 2019, so that the yearly number of diagnoses drops by about 13% between 2018 and 2019, even though in the real data it increases by about 1%:

ma=\(x,b=1,f=b){x[]=rowMeans(embed(c(rep(NA,b),x,rep(NA,f)),f+b+1),na.rm=T);x}

rapid=CJ(year=2018:2025,month=1:12)[!(year==2025&month>5)]
diagnoses=c(24265,23485,24346,25994,27661,26373,25833,25212,24953,24540,24564,25231,25169,24844,24873,26294,26401,25994,25037,25043,24770,25236,24966,25444,25128,25621,24809,16910,17246,18362,19806,21826,23169,23614,24274,24592,23870,23534,24370,26499,27741,25785,25686,25425,25493,25801,25927,25658,25752,26316,25826,27370,28176,27371,26549,26517,26984,26615,27285,26179,26944,26909,26700,28151,29276,27220,26845,26652,26579,26473,28109,27843,27733,27699,28826,28654,29560,28914,28213,27783,27948,27807,27565,28289,27863,28261,27960,28428,27754) # adjusted for number of working days
p1=rapid[,.(x=as.Date(paste(year,month,16,sep="-")),y=diagnoses)]
p1[,y:=ma(y,5,0)]
p1$z="Real data (6-month backward moving average)"

es=CJ(year=2018:2025,month=1:12)[1:87]
es$rate=c(239.6,235.7,237.5,242.7,248.6,250.4,253.2,255.5,256.8,254.2,249.6,247.5,239.3,230.8,222.3,218.2,212.1,206.9,206.1,206.4,206.4,204.6,203.1,202.3,206.1,210.5,215.1,206.4,198.9,191.5,184.5,178.6,176.3,185.1,196.1,204.9,212.1,214.6,216.2,220.8,226.2,227.0,230.3,233.1,234.4,233.7,231.3,230.6,230.6,232.1,232.6,236.0,238.5,241.1,242.4,242.4,244.5,243.4,242.1,240.3,242.1,243.9,244.5,248.1,252.4,255.3,254.5,255.0,254.2,251.7,250.1,250.9,253.5,255.5,260.7,266.1,269.4,272.7,273.5,273.5,271.7,269.1,267.3,266.8,268.4,270.9,272.2)

# nomispop=c(10082761,10243221,10332210,10469044,10630852,10787479,10981092) # exact population estimate of ages 65+
espop=c(55924500,56230100,56326000,56554900,57112500,57690300,57690300,57690300) # from Statista
espop=espop*c(.1803,.1822,.1834,.1851,.186,.1862,.1873,.1873) # proportion aged 65 and above from ONS website

p3=es[,.(x=as.Date(paste(year,month,16,sep="-")),y=rate*espop[factor(year)]/1e5)]
p3$z="Ethical Skeptic's rate converted to number of diagnoses"

p=rbind(p1,p3)[,z:=factor(z,unique(z))]

xstart=as.Date("2018-1-1");xend=as.Date("2025-7-1");xbreak=seq(xstart+182,as.Date("2025-1-1"),"year")
ybreak=pretty(c(0,p$y));ystart=0;yend=max(p$y*1.03)

yearly=p[,.(y=mean(y)),.(year=year(x),z)][year%in%c(2018,2019,2021,2024)]
yearly[,x:=as.Date(paste0(year,"-1-1"))]

color=c(hsv(22/36,1,.6),"#aaaa00")

anno1="ES altered his data to get a 13% decrease between 2018 and 2019, even though the real data has a 1% increase"
anno2="ES altered his data to get a 22% increase between 2021 and 2024, even though the real data only has a 12% increase"
anno=data.table(x=c(xstart+50,xend-50),y=14000,label=c(str_wrap(anno1,25),str_wrap(anno2,34)))

ggplot(p)+
geom_vline(xintercept=seq(xstart,xend,"year"),color="gray90",linewidth=.4)+
annotate("rect",xmin=xstart,xmax=xend,ymin=ystart,ymax=yend,linewidth=.4,lineend="square",linejoin="mitre",fill=NA,color="gray75")+
geom_line(aes(x,y,color=z),linewidth=.6)+
geom_segment(data=yearly,aes(x,y,xend=x+365,yend=y,color=z),linewidth=.6,linetype="42")+
geom_segment(data=yearly[z!=z[1]],aes(x+182+c(100,-100),14500,xend=x+182,yend=y-500),color=color[2],linewidth=.6,arrow=arrow(type="closed",length=unit(4,"pt")),lineend="butt",linejoin="mitre")+
geom_text(data=anno,aes(x,y,label=label),hjust=c(0,1),vjust=1,size=3.87,color=color[2],lineheight=.8)+
labs(x=NULL,y=NULL,title="Monthly cancer diagnoses in England rapid registration data")+
scale_x_continuous(limits=c(xstart,xend),breaks=xbreak,labels=year(xbreak))+
scale_y_continuous(limits=c(ystart,yend),breaks=ybreak,labels=\(x)ifelse(x>=1e3,paste0(x/1e3,"k"),x))+
scale_color_manual(values=color)+
coord_cartesian(clip="off",expand=F)+
guides(color=guide_legend(ncol=1,byrow=F))+
theme(axis.text=element_text(size=11,color="gray40"),
  axis.text.y=element_text(margin=margin(,1.5)),
  axis.ticks=element_line(linewidth=.4,color="gray75"),
  axis.ticks.length=unit(0,"pt"),
  axis.ticks.length.y=unit(4,"pt"),
  legend.background=element_blank(),
  legend.box.spacing=unit(0,"pt"),
  legend.direction="vertical",
  legend.key=element_blank(),
  legend.key.height=unit(12,"pt"),
  legend.key.width=unit(24,"pt"),
  legend.margin=margin(-2,,4),
  legend.position="top",
  legend.spacing.x=unit(2,"pt"),
  legend.spacing.y=unit(0,"pt"),
  legend.text=element_text(size=11),
  legend.title=element_blank(),
  panel.background=element_blank(),
  plot.title=element_text(size=11,hjust=.5,face=2,margin=margin(,,4)))
ggsave("1.png",width=5.1,height=3,dpi=300*4)

I'm fairly sure that ES manually falsified the data. He has not documented his methodology precisely enough that other people could reproduce his work exactly, which also makes it difficult for other people to bust him falsifying the data. But the 13% drop between 2018 and 2019 is completely unrealistic, and it's inconsistent with the real data but consistent with his history of falsifying data.

Some pairs of points in his plot have irregular horizontal spacing, which might be if he somehow manually moved the points around in a GUI, so that he accidentally ended up moving the points horizontally and not only vertically:

From the arrows in the next plot, you can see that I found anomalous features next to two pairs of months that had irregular horizontal spacing. It's also weird how there's several segments of months in 2019 and 2020 that look almost like straight lines, but there are sudden discontinuities in slope before or after the line-like segments, which is anomalous for a plot that is supposed to show a moving average. In my purple line which shows a 6-month moving average of the real data, there are only two clear discontinuities in the slope, which are between March and April of 2020 because the incidence dropped dramatically in April, and between September and October of 2020 because April 2020 fell out of the window of the moving average, but in general my purple line looks much more smooth than Ethical Skeptic's blue line:

ma=\(x,b=1,f=b){x[]=rowMeans(embed(c(rep(NA,b),x,rep(NA,f)),f+b+1),na.rm=T);x}

rapid=CJ(year=2018:2025,month=1:12)[!(year==2025&month>5)]
rapid$diagnoses=c(24265,23485,24346,25994,27661,26373,25833,25212,24953,24540,24564,25231,25169,24844,24873,26294,26401,25994,25037,25043,24770,25236,24966,25444,25128,25621,24809,16910,17246,18362,19806,21826,23169,23614,24274,24592,23870,23534,24370,26499,27741,25785,25686,25425,25493,25801,25927,25658,25752,26316,25826,27370,28176,27371,26549,26517,26984,26615,27285,26179,26944,26909,26700,28151,29276,27220,26845,26652,26579,26473,28109,27843,27733,27699,28826,28654,29560,28914,28213,27783,27948,27807,27565,28289,27863,28261,27960,28428,27754) # adjusted for number of working days

espop=c(55924500,56230100,56326000,56554900,57112500,57690300,57690300,57690300) # from Statista
espop=espop*c(.1803,.1822,.1834,.1851,.186,.1862,.1873,.1873) # proportion aged 65 and above from ONS website

rapid[,x:=as.Date(paste(year,month,1,sep="-"))]
p=rapid[,.(x,y=ma(diagnoses/espop[year-2017]*1e5,5,0))]

xstart=as.Date("2018-1-1");xend=as.Date("2025-7-1");xbreak=seq(xstart,xend-10,"6 month")
xlab=ifelse(month(xbreak)==1,paste0(year(xbreak),"\nJan"),"")
ybreak=pretty(c(0,p$y));ystart=0;yend=max(p$y*1.03)

ystart=120;yend=300;ybreak=0:15*20

ggplot(p)+
geom_point(aes(x,y),size=1.4,stroke=0,color=hsv(5/6,1,.4))+
geom_line(aes(x,y),linewidth=.6,color=hsv(5/6,1,.4))+
labs(x=NULL,y=NULL)+
scale_x_continuous(limits=c(xstart,xend),breaks=xbreak,labels=xlab)+
scale_y_continuous(limits=c(ystart,yend),breaks=ybreak)+
guides(color=guide_legend(ncol=3,byrow=F))+
coord_cartesian(clip="off",expand=F)+
theme(axis.text=element_text(size=9,color="black",lineheight=.8),
  axis.ticks=element_line(color="black",linewidth=.3),
  axis.ticks.length=unit(3,"pt"),
  axis.ticks.x=element_line(color=alpha("black",1:0)),
  legend.background=element_blank(),
  legend.box.spacing=unit(0,"pt"),
  legend.key=element_blank(),
  legend.key.height=unit(12,"pt"),
  legend.key.width=unit(23,"pt"),
  legend.margin=margin(-2,,14),
  legend.position="top",
  legend.spacing.x=unit(2,"pt"),
  legend.spacing.y=unit(0,"pt"),
  legend.text=element_text(size=10,vjust=.5),
  legend.title=element_blank(),
  panel.background=element_blank(),
  panel.grid=element_blank(),
  plot.background=element_rect(fill="transparent",color=NA),
  plot.title=element_text(size=11,hjust=.5))
ggsave("1.png",width=6,height=3,dpi=300*4)

The next plot also shows how the difference between the pre-2020 and post-2021 slopes is greatly exaggerated by Ethical Skeptic's fake data:

ma=\(x,b=1,f=b){x[]=rowMeans(embed(c(rep(NA,b),x,rep(NA,f)),f+b+1),na.rm=T);x}

rapid=CJ(year=2018:2025,month=1:12)[!(year==2025&month>5)]
rapid$cases=c(24265,23485,24346,25994,27661,26373,25833,25212,24953,24540,24564,25231,25169,24844,24873,26294,26401,25994,25037,25043,24770,25236,24966,25444,25128,25621,24809,16910,17246,18362,19806,21826,23169,23614,24274,24592,23870,23534,24370,26499,27741,25785,25686,25425,25493,25801,25927,25658,25752,26316,25826,27370,28176,27371,26549,26517,26984,26615,27285,26179,26944,26909,26700,28151,29276,27220,26845,26652,26579,26473,28109,27843,27733,27699,28826,28654,29560,28914,28213,27783,27948,27807,27565,28289,27863,28261,27960,28428,27754)
p1=rapid[,.(x=as.Date(paste(year,month,16,sep="-")),y=ma(cases,5,0))]
p1$z="Real data (adjusted for working days, 6-month moving average)"

es=CJ(year=2018:2025,month=1:12)[1:87]
es$rate=c(239.6,235.7,237.5,242.7,248.6,250.4,253.2,255.5,256.8,254.2,249.6,247.5,239.3,230.8,222.3,218.2,212.1,206.9,206.1,206.4,206.4,204.6,203.1,202.3,206.1,210.5,215.1,206.4,198.9,191.5,184.5,178.6,176.3,185.1,196.1,204.9,212.1,214.6,216.2,220.8,226.2,227.0,230.3,233.1,234.4,233.7,231.3,230.6,230.6,232.1,232.6,236.0,238.5,241.1,242.4,242.4,244.5,243.4,242.1,240.3,242.1,243.9,244.5,248.1,252.4,255.3,254.5,255.0,254.2,251.7,250.1,250.9,253.5,255.5,260.7,266.1,269.4,272.7,273.5,273.5,271.7,269.1,267.3,266.8,268.4,270.9,272.2)
espop=c(55924500,56230100,56326000,56554900,57112500,57690300,57690300,57690300) # from Statista
espop=espop*c(.1803,.1822,.1834,.1851,.186,.1862,.1873,.1873) # proportion aged 65 and above from ONS website
p2=es[,.(x=as.Date(paste(year,month,16,sep="-")),y=rate*espop[factor(year)]/1e5)]
p2$z="Ethical Skeptic's fake data converted to number of diagnoses"

p=rbind(p1,p2)[,z:=factor(z,unique(z))]

lm=rbind(cbind(p,type=2)[,fit:=year(x)%in%2018:2019],cbind(p,type=3)[,fit:=year(x)%in%2022:2024])
lm=lm[,.(x,y=predict(lm(y~x,.SD[fit==T]),.SD),fit),.(z,type)]
p=rbind(p[,fit:=T][,type:=1],lm)

p[,type:=factor(type,,c("Actual","2018-2019 trend","2022-2024 trend"))]

xstart=as.Date("2018-1-1");xend=as.Date("2025-7-1");xbreak=seq(xstart+182,as.Date("2025-1-1"),"year")
ybreak=pretty(c(0,p$y));ystart=0;yend=max(ybreak)

ggplot(p)+
facet_wrap(~z,ncol=1,dir="v",scales="free_x")+
geom_vline(xintercept=seq(xstart,xend,"year"),color="gray90",linewidth=.4)+
annotate("rect",xmin=xstart,xmax=xend,ymin=ystart,ymax=yend,linewidth=.4,lineend="square",linejoin="mitre",fill=NA,color="gray75")+
geom_line(aes(x,y,color=type),linewidth=.6,linetype="11")+
geom_line(data=p[fit==T],aes(x,y,color=type),linewidth=.6)+
labs(x=NULL,y=NULL,title="Monthly cancer diagnoses in England rapid registration data")+
scale_x_continuous(limits=c(xstart,xend),breaks=xbreak,labels=year(xbreak))+
scale_y_continuous(limits=c(ystart,yend),breaks=ybreak,labels=\(x)ifelse(x>=1e3,paste0(x/1e3,"k"),x))+
scale_color_manual(values=c("black",hsv(1/3,1,.7),hsv(0,.7,1)))+
scale_alpha_manual(values=c(1,.4,.4))+
coord_cartesian(clip="off",expand=F)+
guides(alpha="none")+
theme(axis.text=element_text(size=11,color="gray40"),
  axis.text.y=element_text(margin=margin(,1.5)),
  axis.ticks=element_line(linewidth=.4,color="gray75"),
  axis.ticks.length=unit(0,"pt"),
  axis.ticks.length.y=unit(4,"pt"),
  legend.background=element_blank(),
  legend.box.spacing=unit(0,"pt"),
  legend.key=element_blank(),
  legend.key.height=unit(12,"pt"),
  legend.key.width=unit(24,"pt"),
  legend.margin=margin(-2,,4),
  legend.position="top",
  legend.spacing.x=unit(2,"pt"),
  legend.spacing.y=unit(0,"pt"),
  legend.text=element_text(size=11),
  legend.title=element_blank(),
  panel.background=element_blank(),
  panel.spacing.y=unit(3,"pt"),
  plot.title=element_text(size=11,hjust=.5,face=2,margin=margin(,,4)),
  strip.background=element_rect(fill="gray90",color="gray75",linewidth=.4),
  strip.text=element_text(size=11,margin=margin(3,,3)))
ggsave("1.png",width=5,height=4,dpi=300*4)

In the next plot I calculated age-standardized incidence rates by downloading an age-stratified version of the rapid registration data from here: https://nhsd-ndrs.shinyapps.io/rcrd/ ("Demographic factors > Downloads > Download time trend data for all cancer groups and demographic factors"). I took yearly population estimates by age from Nomis: https://www.nomisweb.co.uk/datasets/pestsyoala. The age-standardized incidence increased by only about 3% between 2022 and 2024:

t=fread("Incidence_Treatment_statistics_England.csv.gz")
t=t[`Cancer group`=="All sites combined"]
t=t[Metric=="New cancer diagnoses (working day adjusted)"]
t=t[Breakdown=="Age-group"]
t[,age:=as.integer(sub("\\D.*","",Demographic))]
t=t[,.(year=Year,month=match(Month,month.name),diagnoses=Statistic,age)]

pop=fread("https://sars2.net/f/englandpop.csv")
ages=c(0,5:8*10)
pop=pop[,.(pop=sum(pop)),.(year,age=ages[findInterval(age,ages)])]
pop=rbind(pop,pop[year==2024][,year:=2025])
t=merge(t,pop)
t=merge(pop[year==2020,.(age,std=pop/sum(pop))],t)

p=t[,.(y=sum(diagnoses/pop*std*1e5)),.(x=as.Date(paste(year,month,16,sep="-")))]

xstart=as.Date("2018-1-1");xend=as.Date("2025-7-1");xbreak=seq(xstart+182,as.Date("2025-1-1"),"year")
ybreak=pretty(c(0,p$y));ystart=0;yend=max(p$y*1.03)

ggplot(p)+
geom_vline(xintercept=seq(xstart,xend,"year"),color="gray90",linewidth=.4)+
annotate("rect",xmin=xstart,xmax=xend,ymin=ystart,ymax=yend,linewidth=.4,lineend="square",linejoin="mitre",fill=NA,color="gray75")+
geom_point(aes(x,y),size=1.3,stroke=0)+
geom_line(aes(x,y),linewidth=.6)+
labs(x=NULL,y=NULL,title="Age-adjusted rate of monthly new cancer diagnoses per\n100,000 people in England rapid registration data")+
scale_x_continuous(limits=c(xstart,xend),breaks=xbreak,labels=year(xbreak))+
scale_y_continuous(limits=c(ystart,yend),breaks=ybreak)+
scale_color_manual(values=c(hsv(0,.7,1),"black"))+
coord_cartesian(clip="off",expand=F)+
guides(color=guide_legend(ncol=1,byrow=F))+
theme(axis.text=element_text(size=11,color="gray40"),
  axis.text.y=element_text(margin=margin(,1)),
  axis.ticks=element_line(linewidth=.4,color="gray75"),
  axis.ticks.length=unit(0,"pt"),
  axis.ticks.length.y=unit(4,"pt"),
  panel.background=element_blank(),
  plot.subtitle=element_text(hjust=.5,margin=margin(,,4)),
  plot.title=element_text(size=11,hjust=.5,face=2,margin=margin(,,4)))
ggsave("1.png",width=5.2,height=3,dpi=300*4)

Without any kind of adjustment for population size or age, the yearly number of diagnoses increased by only about 3% between 2019 and 2022. In Ethical Skeptic's plot, the rate of new diagnoses per population aged 65 and above increased by about 10% between 2019 and 2022, even though his adjustment for population size should make the percentage smaller and not bigger:

From the plot below where I overlaid the yearly data from the NHS report over the rapid registration data, you can see that the rapid registration data is missing about half of all new diagnoses, because it's missing non-melanoma skin cancer which is very common in England, and it's also missing a fraction of diagnoses for other types of cancer:

ma=\(x,b=1,f=b){x[]=rowMeans(embed(c(rep(NA,b),x,rep(NA,f)),f+b+1),na.rm=T);x}

y=fread("https://sars2.net/f/Table_1_machine_readable_updated_June2025.csv.gz")
y=y[stage_at_diagnosis=="All stages"&imd_quintile=="All quintiles"&gender=="Persons"]
y=y[hormone_receptor%in%c("","All")&hormone_receptor_status%in%c("","All")] # `All` is for breast cancer, empty is for other cancers
y=y[ndrs_detailed_group%like%"^All|ALL"]
p1=y[,.(y=sum(count)),.(x=diagnosisyear)][,.(x=as.Date(paste0(x,"-7-1")),y=y/(365+(x%%4==0)))]
p1$z="Cancer Registration Statistics, England, 2022"

rapid=CJ(year=2018:2025,month=1:12)[!(year==2025&month>5)]
rapid$cases=c(24265,23485,24346,25994,27661,26373,25833,25212,24953,24540,24564,25231,25169,24844,24873,26294,26401,25994,25037,25043,24770,25236,24966,25444,25128,25621,24809,16910,17246,18362,19806,21826,23169,23614,24274,24592,23870,23534,24370,26499,27741,25785,25686,25425,25493,25801,25927,25658,25752,26316,25826,27370,28176,27371,26549,26517,26984,26615,27285,26179,26944,26909,26700,28151,29276,27220,26845,26652,26579,26473,28109,27843,27733,27699,28826,28654,29560,28914,28213,27783,27948,27807,27565,28289,27863,28261,27960,28428,27754) # adjusted for number of working days
p2=rapid[,.(x=as.Date(paste(year,month,16,sep="-")),y=cases)][,y:=y/days_in_month(x)]
p2$z="Rapid registration dashboard"

es=CJ(year=2018:2025,month=1:12)[1:87]
es$rate=c(239.6,235.7,237.5,242.7,248.6,250.4,253.2,255.5,256.8,254.2,249.6,247.5,239.3,230.8,222.3,218.2,212.1,206.9,206.1,206.4,206.4,204.6,203.1,202.3,206.1,210.5,215.1,206.4,198.9,191.5,184.5,178.6,176.3,185.1,196.1,204.9,212.1,214.6,216.2,220.8,226.2,227.0,230.3,233.1,234.4,233.7,231.3,230.6,230.6,232.1,232.6,236.0,238.5,241.1,242.4,242.4,244.5,243.4,242.1,240.3,242.1,243.9,244.5,248.1,252.4,255.3,254.5,255.0,254.2,251.7,250.1,250.9,253.5,255.5,260.7,266.1,269.4,272.7,273.5,273.5,271.7,269.1,267.3,266.8,268.4,270.9,272.2)
espop=c(55924500,56230100,56326000,56554900,57112500,57690300,57690300,57690300) # from Statista
espop=espop*c(.1803,.1822,.1834,.1851,.186,.1862,.1873,.1873) # proportion aged 65 and above from ONS website
p3=es[,.(x=as.Date(paste(year,month,16,sep="-")),y=rate*espop[factor(year)]/1e5)]
p3[,y:=y/ma(days_in_month(x),5,0)]
p3$z="Ethical Skeptic's rate converted to number of diagnoses"

p=rbind(p1,p2,p3)[,z:=factor(z,unique(z))]

xstart=as.Date("2013-1-1");xend=as.Date("2025-7-1");xbreak=seq(xstart+182,as.Date("2025-1-1"),"year")
ybreak=pretty(c(0,p$y),7);ystart=0;yend=max(ybreak)

ggplot(p)+
geom_vline(xintercept=seq(xstart,xend,"year"),color="gray90",linewidth=.4)+
annotate("rect",xmin=xstart,xmax=xend,ymin=ystart,ymax=yend,linewidth=.4,lineend="square",linejoin="mitre",fill=NA,color="gray75")+
geom_point(aes(x,y,color=z,alpha=z),size=1.4,stroke=0)+
geom_line(aes(x,y,color=z),linewidth=.6)+
labs(x=NULL,y=NULL,title="New cancer diagnoses in England (divided by number of\ndays in year or month)")+
scale_x_date(limits=c(xstart,xend),breaks=xbreak,labels=year(xbreak))+
scale_y_continuous(limits=c(ystart,yend),breaks=ybreak)+
scale_color_manual(values=c(hsv(1/3,1,.6),hsv(22/36,1,.6),"#aaaa00"))+
scale_alpha_manual(values=c(1,0,0,0))+
coord_cartesian(clip="off",expand=F)+
guides(color=guide_legend(ncol=1,byrow=F))+
theme(axis.text=element_text(size=11,color="gray40"),
  axis.text.y=element_text(margin=margin(,1.5)),
  axis.ticks=element_line(linewidth=.4,color="gray75"),
  axis.ticks.length=unit(0,"pt"),
  axis.ticks.length.y=unit(4,"pt"),
  legend.background=element_blank(),
  legend.box.spacing=unit(0,"pt"),
  legend.direction="vertical",
  legend.key=element_blank(),
  legend.key.height=unit(12,"pt"),
  legend.key.width=unit(24,"pt"),
  legend.margin=margin(-2,,4),
  legend.position="top",
  legend.spacing.x=unit(2,"pt"),
  legend.spacing.y=unit(0,"pt"),
  legend.text=element_text(size=11,margin=margin(,,,1)),
  legend.title=element_blank(),
  panel.background=element_blank(),
  plot.subtitle=element_text(size=11,hjust=.5,margin=margin(,,4)),
  plot.title=element_text(size=11,hjust=.5,face=2,margin=margin(,,4)))
ggsave("1.png",width=5.1,height=3.5,dpi=300*4)

The website of the NHS says: "The Rapid Cancer Registration Data contains proxy tumour registrations and some associated events on the cancer patient pathway (e.g. surgery, radiotherapy and systemic anti-cancer therapy data) from January 2018 to the most recently available data on cancer diagnoses. RCRD provides a quicker, indicative source of cancer data compared to the National Cancer Registration Data (NCRD), which is the 'gold-standard' registration data set and which relies on additional data sources, enhanced follow-up with trusts and expert processing by cancer registration officers. Due to the lower quality of RCRD, the data will not match the eventual National Statistics published on the full NCRD." [https://digital.nhs.uk/ndrs/data/data-sets/rcrd]

Fake linear regression of cancer mortality since 2014

I immediately thought that the slope of his supposed "2014-2019 regression baseline" didn't look steep enough. ES has a habit of faking these baselines or drawing them by hand so that he doesn't do any real regression of the data. When I tried to reproduce his plot, I confirmed that a real linear regression of the data was in fact much steeper than his baseline, as you can see from the red line I added here:

old=fread("https://sars2.net/f/Weekly_Counts_of_Deaths_by_State_and_Select_Causes__2014-2019.csv.gz")
old=old[`Jurisdiction of Occurrence`=="United States"]
old=old[,.(year=`MMWR Year`,week=`MMWR Week`,dead=`Malignant neoplasms (C00-C97)`)]

t=fread("https://sars2.net/f/wondermalignantweekly.csv")[year>2019,.(year,week,dead)]
t=rbind(old,t)

p=t[,.(x=MMWRweek::MMWRweek2Date(year,week,4),y=dead,z="Actual deaths")]
p=rbind(p,p[,.(x,y=predict(lm(y~x,p[year(x)<2020]),p),z="2014-2019 linear regression")])
p[,z:=factor(z,unique(z))]

xstart=as.Date("2014-1-1");xend=as.Date("2025-7-1");xbreak=seq(xstart,xend-1,"6 month")
ystart=10e3;yend=13e3;ybreak=seq(ystart,yend,500)

ggplot(p)+
geom_line(aes(x,y,color=z,linewidth=z))+
labs(x=NULL,y=NULL)+
scale_x_date(limits=c(xstart,xend),breaks=xbreak,labels=ifelse(month(xbreak)==7,year(xbreak),""))+
scale_y_continuous(limits=c(ystart,yend),breaks=ybreak)+
scale_color_manual(values=c("black","#ff6666"))+
scale_linewidth_manual(values=c(.3,.6))+
coord_cartesian(clip="off",expand=F)+
theme(axis.text=element_text(size=9,color="black"),
  axis.text.y=element_blank(),
  axis.ticks=element_line(color="black",linewidth=.3),
  axis.ticks.length=unit(3,"pt"),
  axis.ticks.x=element_line(color=alpha("black",1:0)),
  legend.background=element_blank(),
  legend.box.spacing=unit(0,"pt"),
  legend.justification=c(.5,0),
  legend.key=element_blank(),
  legend.key.height=unit(12,"pt"),
  legend.key.width=unit(23,"pt"),
  legend.position=c(.5,.86),
  legend.spacing.x=unit(2,"pt"),
  legend.spacing.y=unit(0,"pt"),
  legend.text=element_text(size=10,vjust=.5),
  legend.title=element_blank(),
  panel.background=element_blank(),
  panel.grid=element_blank(),
  plot.background=element_rect(fill="transparent",color=NA),
  plot.title=element_text(size=11,hjust=.5))
ggsave("1.png",width=6,height=3.2,dpi=300*4)

Ethical Skeptic's plot says that "Weekly cancer death rates are up above the 2014 - 2019 regression baseline". The way he appears to have accomplished the feat was to first fake his regression line, and then to manually move points upwards when the points happened to fall below his fake regression line.

In the plot above, my black line otherwise matches Ethical Skeptic's blue line, except there are clusters of weeks in 2022 and 2023 where my line is much lower. I think it's because ES manually shifted points up that were below his dashed orange baseline. In 2022 and 2023, his blue line has only 4 points that are below his dashed orange baseline, but my black line has 12 points.

(Ethical Skeptic's blue line is also slightly higher than my black line from 2020 onwards. I took data for 2014-2019 from the same CDC dataset that ES uses in his plots. [https://data.cdc.gov/National-Center-for-Health-Statistics/Weekly-Counts-of-Deaths-by-State-and-Select-Causes/3yf8-kanr] I took data from 2020 onwards from CDC WONDER, but I don't know why my number of deaths from CDC WONDER would be slightly lower than Ethical Skeptic's number of deaths, because I just did a simple query for weekly deaths with the underlying cause malignant neoplasms (C00-C97). So maybe ES applied a very slight upwards adjustment to all points from 2020 onwards.)

Paper about cancer incidence by vaccination status in South Korea

One reason why vaccinated people had a higher incidence of cancer might be that vaccinated people are more likely to get screened for cancer than unvaccinated people, because unvaccinated people tend to have lower health-seeking behavior and to be more hesitant to interact with medical services.

In a Japanese study unvaccinated women were much less likely than vaccinated women to get screening for breast cancer. The authors wrote that when they did a regression adjusted for age, socioeconomic variables, and health-seeking behavior traits, "Individuals who remained unvaccinated due to health concerns (incidence rate ratio (IRR) = 0.47, 95% confidence interval (CI) 0.29-0.77, p = 0.003) and for other unspecified reasons (IRR = 0.73, 95% CI 0.62-0.86, p < 0.001) were significantly less inclined to opt for screening when compared to their fully vaccinated counterparts." [https://www.mdpi.com/2072-6694/16/9/1783]

A Canadian paper from 2024 said: "Unvaccinated respondents were less likely to have received serum cholesterol (aOR 0.69; 95 % CI [0.50-0.70), serum glucose (aOR 0.65; 95 % CI [0.56-0.75]), or blood pressure measurements (aOR 0.47; 95 % CI [0.33-0.66]); and were less likely to have received breast cancer (aOR 0.35; 95 % CI [0.25-0.48]), colorectal cancer (aOR 0.52; 95 % CI [0.46-0.60]) and prostate cancer screening (aOR 0.61; 95 % CI [0.48-0.76])." [https://www.sciencedirect.com/science/article/abs/pii/S0264410X24000070]

A paper about cancer screening behavior in South Korea said: "In the city with the highest cancer screening participation rate, factors such as oral health, physical activity, breakfast habits, and past smoking history were all associated with higher participation. This finding aligns with the conclusion of a previous study that identified a correlation between healthy dietary habits, physical activity motivation, and cancer screening participation [35]. A similar trend has been observed in Lithuania, where factors such as the consumption of fresh vegetables, physical activity, and abstinence from alcohol are associated with higher participation in the national breast cancer screening program for women aged 50-69 [36]. In addition, a previous study of cancer screening activity in adults aged 50 years and older in the United States reported that healthy lifestyles, such as physical activity and smoking cessation, are associated with cancer screening behaviors [27]." [https://www.mdpi.com/2227-9032/13/6/664]

Most of the population of South Korea is vaccinated, so if vaccines would've caused a 27% increase in the incidence of cancer in vaccinated people, you'd expect the overall population of Korea to have a major increase in cancer incidence above the pre-COVID trend. However from this plot of age-standardized cancer incidence in South Korea, you can see that the incidence in 2021 and 2022 roughly fell on the pre-COVID trend: [https://e-crt.org/journal/view.php?doi=10.4143/crt.2025.264]

The Korean study was promoted by Kevin McKernan, who says there has been a massive increase in cancer caused by plasmid DNA from mRNA vaccines (and who blocked me after I showed him why Ethical Skeptic's "SV40 cancer wave" plot was fake): [https://x.com/Kevin_McKernan/status/1972004072439140451]

The authors seem to have referred to AstraZeneca and other viral vector vaccines as "cDNA vaccines". Supplementary file 1 says: "Regarding vaccine types between the first and second vaccinations, 1,928,363 individuals (81.02%) were treated with mRNA vaccines only, 333,698 individuals (14.02%) were given cDNA vaccines only, and 117,967 individuals (4.96%) were administered with heterologous vaccinations in the COVID-19 vaccinated cohort (Table S2)."

Normally the term "DNA vaccine" refers to vaccines that deliver DNA in a plasmid, like the Indian ZyCoV-D vaccine, the Korean Genexine vaccine, and the Inovio vaccine. There is no approved DNA vaccine for COVID in South Korea, even though there have been clinical trials for the Inovio and Genexine DNA vaccines, but they wouldn't account for 14% of total subjects in the cancer study. AstraZeneca and J&J are normally called viral vector vaccines, where a modified adenovirus is used as a vector to deliver spike DNA to the cell. Adenoviruses are DNA viruses, so the spike gene inside the adenovirus vector is coded as DNA that is complementary to the original RNA, so the term "cDNA vaccine" was likely used in the paper as a nonstandard term for a viral vector vaccine.

A paper about COVID vaccination in South Korea said: "As of September 25, 2022, 128,710,064 doses of COVID-19 vaccines were administered: BNT162b2 (62.9%), mRNA-1273 (19.5%), ChAdOx1 (15.8%), Ad26.COV2.S (1.2%), NVX-CoV2373 (0.6%) and GBP510 (< 0.1%)." [https://jkms.kr/DOIx.php?id=10.3346%2Fjkms.2022.37.e351] So if AstraZeneca accounted for about 16% of doses in South Korea, and about 14% of subjects in the cancer study were in the cDNA-only group, then it's likely that the cDNA group included AstraZeneca.

Table S3 has a list of vaccine brands that were administered for booster doses, which includes only the three brands of Pfizer, Moderna, and AstraZeneca. There's only a few doses of AstraZeneca listed in the table, because AstraZeneca was generally only administered for primary course doses. The distribution of vaccine brands for primary course doses was not shown anywhere in the paper.

Cancers take time to develop, but Figure 1B shows that even during the first month of follow-up, vaccinated people already had a much higher incidence of cancer than unvaccinated people:

Supplementary file 2 says that the follow-up period started from the day after the primary series was completed, and not from the day of the first dose. But still, most people got the second dose within a few weeks from the first dose, so if the difference in incidence during the first month of follow-up would be explained by vaccines causing cancer, then the vaccines would have to be causing super fast-acting turbo cancer.

Uncle John Returns posted the images below and wrote: "Why should cancer registration rates among the unvaccinated approximately double in a year? Unless they were undercounted to begin with?" [https://x.com/UncleJo46902375/status/1972676088640757796]

His plot demonstrates clearly how unvaccinated people had low incidence during the first 3 months of follow-up, even though you can ignore his "South Korea average 2022" line, because it is not too meaningful. The line shows the age-standardized incidence of cancer in 2022 among the overall population of Korea, which was about 522.7 cases per 100,000 person-years, which he converted to about 4.3 cases per 10,000 people per a 30-day period. [https://www.cancerdata.re.kr/surveillance/en/data?menuId=40]

However Uncle John didn't take into account that the Korean study excluded ages below 20, and in general it doesn't make too much sense to compare the incidence among the PSM-matched cohorts against the incidence among the overall population of Korea.

The authors of the Korean study did propensity score matching to match vaccinated people to unvaccinated people 4-to-1, which ended up skewing the age distribution of the cohorts included in the study, because more than 80% of elderly people were vaccinated, so many vaccinated elderly people had to be discarded. The age distribution of the subjects before the PSM was not shown anywhere, but Table S2 shows the distribution after the PSM, which is compared here to the age distribution of the overall Korean population: [https://x.com/UncleJo46902375/status/1972945872452517945]

The PSM similarly also skewed other attributes of the matched cohorts relative to the distribution of the attributes among the overall population of Korea.

And the Korean study also excluded non-melanoma skin cancer (C44), which was not excluded in the dataset Uncle John used to calculate the "South Korea average 2022" line.

But apparently the matching was done based on the insurance level, and the income level was estimated from the insurance level. Supplementary file 2 says: "The following covariates were considered: age, sex, insurance levels, Charlson comorbidity index (CCI) scores, and prior COVID-19 infection (history of SARS-CoV-2 infection). Insurance levels (recipients of medical aid, grades 1-5, grades 6-10, grades 11-15, and grades 16-20) were defined based on the National Health Insurance premium, which was used as a proxy for income since it is proportional to monthly income and includes both earnings and capital gains." [https://static-content.springer.com/esm/art%3A10.1186%2Fs40364-025-00831-w/MediaObjects/40364_2025_831_MOESM2_ESM.docx]

But anyway, the cohorts were matched by so few variables that there likely remains major residual confounding for the level of health-seeking behavior.

I initially thought that unvaccinated people were assigned index dates that matched the index dates of the paired vaccinated people. But supplementary file 2 said that the index date of unvaccinated people was always set as January 1st 2022: "For the unvaccinated group, the index date was set as January 1, 2022; those who had a vaccination history within 1 year based on the index date and those who were deceased were excluded (n = 30,955). A total of 599,124 unvaccinated individuals were included. For the vaccinated group, the index date was set as the day after the vaccination completion date, and those with incomplete vaccination (n = 278,610), unspecified vaccine type (n = 77,674), deceased (n = 33,836), and a prior medical history of overall cancers within 1-year based on index date (n = 499,572) were excluded."

So the follow-up period of all unvaccinated people consisted of January 1st 2022 to December 31st 2022, but the follow-up period of most vaccinated people already started in 2021.

So the reason why unvaccinated people had low incidence during the first 3 months may have been if people avoided getting screening during COVID waves. The first two major COVID waves in South Korea were the Delta wave which peaked around December 2021, and the Omicron wave which peaked around March 2022:

kim=\(x)ifelse(x>=1e3,ifelse(x>=1e6,paste0(x/1e6,"M"),paste0(x/1e3,"k")),x)

t=fread("https://srhdpeuwpubsa.blob.core.windows.net/whdh/COVID/WHO-COVID-19-global-data.csv")

p=t[Country_code=="KR"][,.(x=Date_reported,dead=New_deaths,case=New_cases)]

xstart=as.Date("2020-1-1");xend=as.Date("2024-1-1");xbreak=seq(xstart+182,xend,"year")
ybreak=0:6*5e5;ystart=0;yend=max(ybreak);ybreak2=ybreak/1e3
secmult=1e3

color=c("black",hsv(0,.5,1))

lab=p[,.(lab=c("Cases","Deaths"),x=c(xstart+50,xend-50),y=yend*.92)]

ggplot(p)+
geom_vline(xintercept=seq(xstart,xend,"year"),color="gray90",linewidth=.4)+
geom_segment(data=CJ(x=seq(xstart,xend,"month")),aes(x,0,xend=x,yend=yend*.015),color="gray75",linewidth=.4)+
geom_hline(yintercept=ybreak,color="gray90",linewidth=.4)+
annotate("rect",xmin=xstart,xmax=xend,ymin=ystart,ymax=yend,linewidth=.4,lineend="square",linejoin="mitre",fill=NA,color="gray75")+
geom_line(aes(x,case),linewidth=.6)+
geom_line(aes(x,dead*secmult),color=color[2],linewidth=.6)+
geom_text(data=lab,aes(x,y,label=lab,hjust=0:1),color=color,size=3.87)+
labs(x=NULL,y=NULL,title="South Korea: Weekly COVID cases and deaths by date reported to WHO",subtitle="Source: data.who.int/dashboards/covid19/data")+
scale_x_date(limits=c(xstart,xend),breaks=xbreak,labels=year(xbreak))+
scale_y_continuous(limits=range(ybreak),labels=kim,breaks=ybreak,sec.axis=sec_axis(trans=~./secmult,breaks=ybreak2,labels=kim))+
coord_cartesian(clip="off",expand=F)+
theme(axis.text=element_text(size=11,color="gray40"),
  axis.text.y=element_text(margin=margin(,2,,2)),
  axis.text.y.left=element_text(color=color[1]),
  axis.text.y.right=element_text(color=color[2]),
  axis.ticks.length=unit(0,"pt"),
  legend.background=element_rect(color="gray70",linewidth=.4),
  legend.box.spacing=unit(0,"pt"),
  legend.direction="vertical",
  legend.justification=c(.5,.5),
  legend.key=element_blank(),
  legend.key.height=unit(12,"pt"),
  legend.key.width=unit(24,"pt"),
  legend.margin=margin(3,5,3,3),
  legend.position=c(.5,.5),
  legend.spacing.x=unit(2,"pt"),
  legend.spacing.y=unit(0,"pt"),
  legend.text=element_text(size=11),
  legend.title=element_blank(),
  panel.background=element_blank(),
  plot.subtitle=element_text(hjust=.5,margin=margin(,,4)),
  plot.title=element_text(size=11,hjust=.5,face=2,margin=margin(,,4)))
ggsave("1.png",width=5.5,height=3.2,dpi=300*4)

I didn't find monthly cancer incidence data for 2022, so I wasn't able verify if the cancer incidence was low in early 2022 relative to later in 2022.

But for example an article from April 2022 said in Korean: "As the spread of COVID-19 continues, there are many people who are procrastinating the check-up, so please do not delay and actively participate in the early check-up as there is a concern about congestion due to the year-end concentration." [https://blog.naver.com/hdnews9001/222715512561] In the Korean cancer screening program, people born on even years are asked to get screening during even years, and people born on odd years are asked to get screening during odd years. People have until the end of the year to get screening, but apparently due to COVID, some people delayed getting screening until the end of the year.

The deadline for screening in 2021 was extended until June 2022 due to COVID. An article from January 2022 said in Korean: "2021 flew by in the blink of an eye! Many people may have postponed their national health screenings to adhere to quarantine guidelines and then forgot about them. In December 2021, the government announced, 'To ensure access to screenings for those who have postponed or refrained from using them to comply with COVID-19 guidelines, the health screening period will be extended.'" [https://blog.ibk.co.kr/2703]

I didn't find data for overall monthly cancer screening rates, but Uncle John Returns found a paper with the monthly number of colonoscopies performed in 2019-2021, where each year there were more colonoscopies performed towards the end of the year than the beginning of the year (so even in 2019, people seem to have procrastinated until the end of the year to get screening). Data for the Omicron wave in spring 2022 is missing, but there was a big drop in colonoscopies performed during the first COVID scare around March 2020: [https://www.irjournal.org/journal/view.php?number=1091]

I now found that screening volume was higher than normal in 2021. The screening volume was much lower than normal in 2022 for breast cancer, but close to the pre-COVID trend for stomach cancer, colorectal cancer, and cervical cancer: [https://www.cancerdata.re.kr/surveillance/en/data?menuId=38]

The follow-up period of all unvaccinated people started from January 1st 2022, so 100% of unvaccinated follow-up time was in 2022, but vaccinated follow-up time was divided between 2021 and 2022, which might have skewed the results of the study.

Was the median age at death from COVID 79.5 years in 2025?

Ethical Skeptic retrieved the number of deaths with UCD COVID in 2025 by 10-year age groups, and since the median group was 75-84, he said that the median age at death was 79.5, which was the average of the minimum and maximum age within the age group: [https://x.com/EthicalSkeptic/status/1971310148619522151]

He could've calculated the exact median age if he would've used broad age groups for ages with suppressed deaths and single-year age groups for ages with no suppressed deaths.

When I did a query at CDC WONDER on September 30th 2025 UTC, there were a total of 10,470 deaths in 2025 with the underlying cause COVID. CDC WONDER suppresses the number of deaths when there are 1 to 9 deaths, but there were 10 or more deaths for each single year of age from 40 up to 99, and even for ages 100 and above. Ages 0-39 had 141 deaths. The median age of death was 82:

The mean age was about 79.3 when I treated ages 0-39 as age 0, or about 79.9 when I treated ages 0-39 as age 39:

The median age of COVID deaths seems to have gone up over time, because it was 72 in 2021, 77 in 2022, and 81 in 2023. The median age was particularly low in 2021, because in 2021 there were still many unvaccinated people who had not acquired natural immunity, and unvaccinated people are younger than vaccinated people:

Ethical Skeptic's image says: "Current U.S. life expectancy for the average 80-year old = 78.4 years". I don't know if he made a typo, because the life expectancy at age 80 should be more than 80 (or in case he meant remaining life expectancy, it should be much lower than 78.4). In the 2023 US life table, the remaining life expectancy for age 80 is about 9.3 years. [https://ftp.cdc.gov/pub/Health_Statistics/NCHS/Publications/NVSR/74-06/Table01.xlsx]

When I took the number of deaths in 2023 from CDC WONDER and I took mid-year resident population estimates for 2023 from the vintage 2024 release by the Census Bureau, I also got a remaining life expectancy of about 9.3 years at age 80:

Comments to Ethical Skeptic (part 4) - sars2.net

Contents

Cancer diagnoses in England per population aged 65 and above

Fake linear regression of cancer mortality since 2014

Paper about cancer incidence by vaccination status in South Korea

Was the median age at death from COVID 79.5 years in 2025?