1
CORRESPONDENCE
I econcilable S a is ical Pa adoxes Ques ion he Validi y o Kim HJ e al.'s S udy on
COVID-19 Vaccina ion and Psychia ic Ou comes (Mol Psychia y 2024; 29: 3635–3643)
Ma co Rocce i
Depa men o Compu e Science and Enginee ing
Uni e si y o Bologna, 40126, I aly
ma co. [email p o ec ed]
Abs ac : The s udy by Kim e al. [1] on psychia ic ou comes ollowing COVID-19 accina ion is s a is ically
in alid due o a se e e, unco ec ed selec ion bias. This me hodological law yields i econcilable epidemiological
pa adoxes, including an implausible ≈ 77% isk educ ion o Schizoph enia (i.e., 1 - HR=0.231) in he high-
isk g oup, and s a is ically impossible incidence a es o Bipola and Anxie y Diso de s, inconsis en wi h obus
Ko ean na ional benchma ks [2-4]. The simul aneous occu ence o HR ≪1 and HR ≫ 1 ac oss di e en
pa hologies p o es he obse ed e ec is a ma hema ical a i ac o misclassi ica ion bias. A alid e-analysis
using obus me hods like P opensi y Sco e Ma ching [5] is wa an ed.
We ead wi h conce n he s udy by Kim HJ e al on he associa ion be ween COVID-19 accina ion and he
incidence o psychia ic diso de s in he Sou h Ko ean Na ional Heal h Insu ance Se ice (NHIS) coho [1].
Gi en he c i ical na u e o da a conce ning na ional and in e na ional heal h policies ega ding COVID-19 and
accina ions, he p oduc ion o s a is ically unsound esul s can pose a signi ican isk o public heal h by
misin o ming decision-making. The e o e, hese laws mus be co ec ed, e en a yea a e publica ion. Ou
analysis concludes ha he s udy’s p ima y indings a e s a is ically in alid and d i en by a se e e, unco ec ed
selec ion bias. This law mani es s in a se ies o highly implausible Haza d Ra ios (HRs) and incidence a es ha
a e i econcilable wi h es ablished Sou h Ko ean epidemiological da a, as epo ed in [2-5].
2
The mos s iking s a is ical anomaly o [1] is he epo ed alse p o ec i e e ec HR = 0.231 (95% CI: 0.164–
0.326) o Schizoph enia (ICD-10 F20-F29) (see Abs ac o [1]). The accina ed g oup is ≈10.5 yea s olde and
signi ican ly less heal hy a baseline (sou ced om Table 1 o [1]). His o ical da a om he Ko ean Na ional
Heal h Insu ance da abase shows an annual incidence a e o Schizoph enia o [77.6 - 88.5] pe 100,000
indi iduals (see Abs ac o [4]). This ange ansla es o a qua e ly incidence o app oxima ely 2.0 - 2.2 pe
10,000 popula ion (e.g., (88.5/10) /4 ≈ 2.21). This ange is consis en wi h he epo ed un accina ed a e 1.98
pe 10,000 o [1] (Sec ion i led: The cumula i e incidences pe 10,000 o psychia ic AEs ollowing he COVID-
19 accina ion). Ins ead, he cumula i e incidence in he accina ed, high- isk g oup 0.51 pe 10,000 (same
Sec ion o [1] as abo e) is nea ly ou imes lowe han bo h he con ol g oup's a e and he his o ically expec ed
a e. Hence, i is scien i ically implausible o he incidence o a ch onic diso de o be so signi ican ly lowe in
he olde , less heal hy g oup compa ed o he younge , heal hie con ol. This pa e n is he i s signa u e o a
misclassi ica ion bias: ob iously, ch onic Schizoph enia pa ien s in he high- isk accina ed g oup we e
sys ema ically excluded o misca ego ized as "no new cases", a i icially lowe ing he nume a o and in alida ing
he HR.
Fu he mo e, he incidence a e epo ed o Bipola Diso de (BD) in he con ol g oup is s a is ically impossible
when benchma ked agains Sou h Ko ean na ional epidemiological da a. The incidence o BD in he un accina ed
g oup was gi en as 1.39 pe 10,000 o e h ee mon hs in [1] (Sec ion i led: The cumula i e incidences pe 10,000
o psychia ic AEs ollowing he COVID-19 accina ion). This a e, howe e , exceeds he epo ed na ional
p e alence ( o al exis ing cases) o Bo de line Pe sonali y Diso de (a se e e, ch onic condi ion o en como bid
wi h BD) which was 1.06 pe 10,000 o e an en i e yea (Resul s Sec ion o [2]). As a undamen al
epidemiological ule, in ac , he incidence o a se e e ch onic condi ion o e h ee mon hs canno exceed he
annual p e alence o a simila ch onic condi ion in he same popula ion. This absolu e inconsis ency con i ms
ha he au ho s' me hodology o de ining "new cases" o BD is undamen ally lawed, ende ing he calcula ed
HR o 0.672 (Abs ac o [1]) meaningless.
The selec ion bias o [1] c ea es alsely low HRs o ch onic diseases (Schizoph enia, BD) and simul aneously
c ea es alsely high HRs o common diseases like Anxie y, Dissocia i e, S ess-Rela ed, and Soma o o m
Diso de s (F40-F48). The esul ing HR o 1.439 [95% CI] = 1.322–1.568 (see Abs ac o [1]) is a di ec
3
ma hema ical a i ac o he unco ec ed baseline dispa i y. The incidence in he accina ed g oup (28.41 pe
10,000) is epo ed as ≈ 40% highe (28.41/20.27 ≈ 1.40) han he con ol g oup (20.27 pe 10,000). All hese
igu es a e sou ced om [1] (Sec ion, i led: The cumula i e incidences pe 10,000 o psychia ic AEs ollowing
he COVID-19 accina ion). The unco ec ed baseline dispa i y is demons a ed by compa ing he s udy's a es o
obus na ional epidemiology. The 12-mon h p e alence o Anxie y Diso de s in Ko ea is 3.1% (Sec ion Anxie y
in [3]), which ansla es o an annual a e o 310 pe 10,000. The incidence o a ch onic condi ion (new cases
appea ing o e ime) mus , by de ini ion, be lowe han i s o al p e alence (cases exis ing a a poin in ime).
The e o e, he qua e ly ac ion o he na ional p e alence, ≈ 77.5 pe 10,000 (i.e., 310/4), ep esen s an absolu e
heo e ical uppe bound o any plausible h ee-mon h incidence a e. The ac ha bo h he accina ed a e (28.41)
and he non- accina ed a e (20.27) a e massi ely lowe han his expec ed na ional uppe bound (77.5) p o es
he exis ence o se e e misclassi ica ion bias in bo h coho s ( accina ed and non- accina ed). Indeed, his s udy
[1] ails again o cap u e he as majo i y o new cases. This me hodological ailu e ende s he de i ed HR
meaningless.
No only bu he simul aneous p esence o HR ≪ 1 (sugges ing a mi aculous p o ec i e e ec on ch onic
diso de s highly p e alen in he accina ed g oup) and HR ≫1 (sugges ing a s ong de imen al e ec on
common diso de s linked o he unco ec ed selec ion bias) wi hin he same non-co ec ed coho is he de ini i e
p oo ha he obse ed e ec is no biological, bu he shadow o he unde lying bias, e lec ing di e en ly ac oss
dis inc pa hologies.
In closing, he h ee s a is ical pa adoxes we ha e exposed demons a e ha he indings o [1]. a e un eliable.
Ou inal in o med opinion is ha he co e p oblem lies in he cons uc ion o he con ol coho : he au ho s s a e
ha he un accina ed g oup was c ea ed by selec ing 50% o un accina ed indi iduals a andom om he gene al
popula ion (see Abs ac o 1). This me hod, while seemingly andom, is s a is ically inadequa e in a e ospec i e
s udy u ilizing adminis a i e da a, as i ails o co ec o he as , in insic di e ences, such as age, como bidi ies,
and heal h-seeking beha io , be ween hose who chose o be accina ed and hose who did no . The obse ed
anomalies a e he di ec consequence o his me hodological e o . We he e o e sugges ha any alid e-analysis
4
o his da a mus employ a s a is ically obus p ocedu e, such as P opensi y Sco e Ma ching (PSM) [5], o achie e
ue co a ia e balance and o ob ain eliable es ima es.
Re e ences
1. Kim HJ, Kim MH, Choi MG, Chun EM. Psychia ic ad e se e en s ollowing COVID-19 accina ion: a
popula ion-based coho s udy in Seoul, Sou h Ko ea. Mol Psychia y. 2024;29(12):3635–3643. doi:
10.1038/s41380-024-02627-0
2. Shin H, Lee HS, Lee BC, Pa k G, U anbileg K. The P e alence and Clinical Cha ac e is ics o Bo de line
Pe sonali y Diso de in Sou h Ko ea Using Na ional Heal h Insu ance Se ice Cus omized Da abase.
Yonsei Med J. 2023;64(9):566–72. doi: 10.3349/ymj.2023.0071.
3. Rim SJ, Hahm B-J, Seong SJ, Pa k JE, Chang SM, Kim B-S, e al. P e alence o Men al Diso de s and
Associa ed Fac o s in Ko ean Adul s: Na ional Men al Heal h Su ey o Ko ea 2021. Psychia y In es ig.
2023;20(3):262–272. doi:10.30773/pi.2022.0307
4. Cho SJ, Kim J, Kang YJ, Lee SY, Seo HY, Pa k JE, e al. Annual P e alence and Incidence o
Schizoph enia and Simila Psycho ic Diso de s in he Republic o Ko ea: A Na ional Heal h Insu ance
Da a-Based S udy. Psychia y In es ig. 2020 Jan 25;17(1):61–70. doi. 10.30773/pi.2019.0041
5. D’Agos ino RB J . P opensi y sco e me hods o bias educ ion in he compa ison o a ea men o a
non- andomized con ol g oup. S a Med. 1998;17(19):2265–81. doi: 10.1002/(sici)1097-
0258(19981015)17:19<2265::aid-sim918>3.0.co;2-b
Au ho In o ma ion
Ma co Rocce i: Depa men o Compu e Science and Enginee ing, Uni e si y o Bologna, 4016 Bologna, I aly,
ma co. occe [email protected] . ORCID: 0000-0003-1264-8595
Con ibu ions
MR conduc ed he da a analysis, concep ualized he s a is ical a gumen s, and w o e he pape .
Co esponding au ho
Co espondence o Ma co Rocce i, ma co. occe [email protected] .
5
E hics decla a ions:
Compe ing in e es s
The au ho decla es no compe ing in e es s.
Funding
This esea ch ecei ed no ex e nal unding
Da a A ailabili y
The da a p esen ed he e is ei he included di ec ly o was ex ac ed om he e e enced documen s. All
calcula ions a e easily ep oducible based on he de ini ions p o ided
E hics app o al and consen o pa icipa e
This s udy uses publicly a ailable, agg ega ed da a ha con ains no p i a e in o ma ion. The e o e, e hical
app o al is no equi ed
