ADHD 的連續配置空間:從亞臨床特徵到臨床診斷
英文題名: The Continuous Configuration Space of ADHD: From Subthreshold Traits to Clinical Diagnosis系列: ADHD 動態配置與認知拓撲系列,第 8 篇版本: v0.1日期: 2026-08-17作者: Neo.K(許筌崴)協作: GPT-5.6 Sol文件性質: 理論建模/認知科學命題/研究綱領文獻檢索截點: 2026-08-17
0. 醫學、分類與證據邊界聲明
本文不是臨床研究、診斷工具、治療指南、流行病學研究或醫療建議。
本文提出的「連續配置空間」「診斷決策區域」「亞臨床配置」「局部聚類」「配置距離」「功能損害流形」等概念均屬待驗證理論命題,不代表已被醫學界確認的 ADHD 病理機制或正式診斷模型。
原作者並非醫學、精神醫學、神經科學、遺傳學或臨床心理專業研究者。本文不提供新的臨床、人體、遺傳、神經影像、流行病學或心理實驗數據;所有實證性背景均來自公開同行評審研究與正式分類資料。
本文不得被用於:
自行診斷 ADHD;
將「亞臨床特徵」直接等同正式 ADHD;
將未達診斷標準者稱為「隱藏 ADHD」;
以數學距離取代專業臨床判斷;
推翻或修改既有個人醫療診斷。
本文最重要的邊界為:
ADHD-like traits ≠ clinical ADHD diagnosis . \boxed{
\text{ADHD-like traits}
\neq
\text{clinical ADHD diagnosis}.
} ADHD-like traits = clinical ADHD diagnosis .
以及:
continuous liability ≠ absence of clinically useful categories . \boxed{
\text{continuous liability}
\neq
\text{absence of clinically useful categories}.
} continuous liability = absence of clinically useful categories .
一個疾病或症候群即使建立在連續風險與症狀分布上,臨床仍可以基於功能損害、持續性、跨情境性、發展史、鑑別診斷與治療需要建立有實用價值的類別判定。
核心限制:
dimensional ontology ≠ automatic diagnostic equivalence . \boxed{
\text{dimensional ontology}
\neq
\text{automatic diagnostic equivalence}.
} dimensional ontology = automatic diagnostic equivalence .
摘要
ADHD 長期同時具有兩種看似競爭的描述。臨床分類需要回答「是否符合 ADHD 診斷」,因此使用類別式決策;人口、遺傳、症狀與生活品質研究則反覆顯示,inattention、hyperactivity/impulsivity 與相關功能困難廣泛分布於一般人口中,且未必存在自然、陡峭的生物學或功能斷點。
2024 年 Arildskov 等人在 1,967 名一般人口學童中檢驗 ADHD traits 與 psychosocial quality of life 的關係,未發現高 trait 區域存在突然的 QoL 崩落門檻,結果更接近線性下降。2025 年 van der Laan 等人的大型 genome-wide association meta-analysis 整合 290,134 次 ADHD symptom measures、70,953 名獨立個體與 ADHD diagnosis GWAS,結論支持 clinical ADHD 位於由 ADHD symptoms 索引的 continuous liability 高端。2025 年 Knyspel 等人分析 10,454 名 21 歲雙生子,psychometric bifactor model 支持 ADHD symptoms 的 general dimension,同時保留 inattention 與 hyperactivity 的 secondary dimensions;遺傳層面又顯示這些維度不能簡化為單一完全同質因素。
然而,連續性並不排除局部結構。2026 年 Pan 等人的 JAMA Psychiatry 多中心兒童研究,在 normative morphometric similarity network deviations 上辨識出三個可外部驗證的 ADHD biotypes,並明確指出其結果同時提供 dimensional 與 categorical insights。2024 年 Nature Genetics 的 14,084 名已診斷 ADHD 個案研究亦顯示,不同共病與診斷歷程群體具有不同 polygenic profiles,支持 ADHD 內部存在遺傳異質性。
基於上述證據,本文提出「Continuous Configuration Space Hypothesis, CCSH」。令個體在時間 t t t 的 ADHD-related configuration 為:
c i , t ∈ Ω C ⊆ R d . \mathbf c_{i,t}
\in
\Omega_C
\subseteq
\mathbb R^d. c i , t ∈ Ω C ⊆ R d .
其中 c i , t \mathbf c_{i,t} c i , t 不代表診斷,而是由注意配置、抑制、工作記憶、喚醒、獎勵、情緒調節、切換、時間組織、情境敏感度等候選維度構成的中層狀態向量。
臨床診斷則表示為另一層決策算子:
D i , t = Γ ( y i , t , I i , t , H i , X i , t ) , \mathcal D_{i,t}
=
\Gamma
\left(
\mathbf y_{i,t},
I_{i,t},
H_i,
X_{i,t}
\right), D i , t = Γ ( y i , t , I i , t , H i , X i , t ) ,
其中 y \mathbf y y 為可觀察症狀, I I I 為功能損害, H H H 為發展史, X X X 為跨情境與鑑別診斷等臨床資訊。
因此:
c can be continuous while D is categorical . \boxed{
\mathbf c
\text{ can be continuous while }
\mathcal D
\text{ is categorical}.
} c can be continuous while D is categorical .
本文定義「subthreshold/亞臨床」不是「其實已經有 ADHD」,而是:
Γ ( ⋅ ) = 0 \Gamma(\cdot)=0 Γ ( ⋅ ) = 0
的前提下,仍可能存在部分 ADHD-related traits 或功能困難。這些個體可能需要支持,但支持需要不等於 ADHD 診斷成立。
本文進一步提出:高維配置空間可以同時具有連續密度、局部聚類、不同功能損害曲面與任務依賴邊界;臨床 diagnosis 是在此空間之上建立的決策層,而不是自然界必然存在的一條單一歐氏幾何切線。
最終可證偽問題為:
Does a high-dimensional continuous-plus-cluster model predict impairment, course, and treatment-relevant outcomes better than either a pure binary model or a pure single-axis spectrum? \boxed{
\text{Does a high-dimensional continuous-plus-cluster model
predict impairment, course, and treatment-relevant outcomes
better than either a pure binary model or a pure single-axis spectrum?}
} Does a high-dimensional continuous-plus-cluster model predict impairment, course, and treatment-relevant outcomes better than either a pure binary model or a pure single-axis spectrum?
關鍵詞: ADHD、dimensional model、categorical diagnosis、subthreshold、continuous liability、biotype、heterogeneity、functional impairment、polygenic risk、configuration space
1. 問題:ADHD 到底是「類別」還是「光譜」?
這個問題若只允許:
Category ∨ Dimension , \text{Category}
\lor
\text{Dimension}, Category ∨ Dimension ,
本身可能就是錯誤二分。
因為我們其實至少在問三個不同問題。
第一:
症狀與風險在一般人口中如何分布?
第二:
生物、認知或行為資料中是否存在可重現的局部群集?
第三:
臨床上何時應建立一個可操作的診斷決策?
這三個問題可以得到不同答案。
因此:
population distribution ≠ latent clustering ≠ clinical classification . \boxed{
\text{population distribution}
\neq
\text{latent clustering}
\neq
\text{clinical classification}.
} population distribution = latent clustering = clinical classification .
2. 連續性與類別性可以同時成立
假設某個底層 liability:
z ∈ R . z
\in
\mathbb R. z ∈ R .
它可以連續分布:
z ∼ p ( z ) . z
\sim
p(z). z ∼ p ( z ) .
臨床卻仍可設定:
z > θ z>\theta z > θ
作為某一決策規則的一部分。
因此:
z continuous ⇏ clinical categories meaningless . \boxed{
z\text{ continuous}
\not\Rightarrow
\text{clinical categories meaningless}.
} z continuous ⇒ clinical categories meaningless .
同樣:
diagnostic category useful ⇏ z naturally binary . \boxed{
\text{diagnostic category useful}
\not\Rightarrow
z\text{ naturally binary}.
} diagnostic category useful ⇒ z naturally binary .
3. 2024 QoL 研究:沒有看到自然跳變點
Arildskov 等人研究:
N = 1967 N=1967 N = 1967
名 6–11 歲一般人口學童。
研究檢驗:
ADHD traits → psychosocial QoL \text{ADHD traits}
\rightarrow
\text{psychosocial QoL} ADHD traits → psychosocial QoL
是否在高 trait 區突然出現明顯非線性下降。
結果沒有找到:
θ natural \theta_{\text{natural}} θ natural
式的明顯 QoL 跳變點。
較符合:
trait severity ↑ ⇒ QoL ↓ \text{trait severity}\uparrow
\Rightarrow
\text{QoL}\downarrow trait severity ↑⇒ QoL ↓
的漸進關係。
因此:
functional burden may be dimensional . \boxed{
\text{functional burden may be dimensional}.
} functional burden may be dimensional .
但這不表示:
diagnostic threshold \text{diagnostic threshold} diagnostic threshold
毫無臨床功能。
4. 症狀門檻與損害門檻不是同一件事
早期 impairment research 已指出:
symptom severity \text{symptom severity} symptom severity
與:
functional impairment \text{functional impairment} functional impairment
彼此相關但不相同。
因此:
S symptom ≠ I impairment . \boxed{
S_{\text{symptom}}
\neq
I_{\text{impairment}}.
} S symptom = I impairment .
一個人可以:
S symptom ↑ S_{\text{symptom}}\uparrow S symptom ↑
但當下:
I impairment I_{\text{impairment}} I impairment
受到環境支架保護。
反過來也可能有:
S symptom S_{\text{symptom}} S symptom
不極端,
但:
I impairment ↑ I_{\text{impairment}}\uparrow I impairment ↑
因為任務與環境要求高度不匹配。
5. 成人 ADHD 的 impairment assessment 本身仍在發展
2024 年 Fuermaier 等人調查 92 名參與 ADHD clinical practice/research 的國際專業人士。
研究發現:
成人 ADHD impairment measurement 使用工具高度多樣;
部分工具的 psychometric properties 可能不足;
臨床與研究者對現行實務存在不滿;
仍需要更好的 impairment assessment。
因此:
impairment is clinically essential but not trivially measured . \boxed{
\text{impairment}
\text{ is clinically essential but not trivially measured}.
} impairment is clinically essential but not trivially measured .
這使任何以單一 symptom cutoff 取代完整功能評估的理論都過度簡化。
6. 2025 Genetics:Clinical ADHD 位於 Continuous Liability 高端
2025 年 van der Laan 等人進行大型 GWAS meta-analysis。
其 symptom data 包含:
290 , 134 290,134 290 , 134
次 ADHD symptom measurements,
來自:
70 , 953 70,953 70 , 953
名獨立個體,
涵蓋多個:
raters;
ages;
instruments。
研究再與 ADHD diagnosis GWAS 整合。
其整體結果支持:
clinical ADHD lies toward the extreme of a continuous liability indexed by ADHD symptoms . \boxed{
\text{clinical ADHD lies toward the extreme
of a continuous liability indexed by ADHD symptoms}.
} clinical ADHD lies toward the extreme of a continuous liability indexed by ADHD symptoms .
這是一個重要的遺傳層支持。
但:
genetic continuity \text{genetic continuity} genetic continuity
不表示:
one-gene-one-axis model . \text{one-gene-one-axis model}. one-gene-one-axis model .
7. Polygenic 不等於單維
若疾病風險來自許多 genetic variants:
G = ∑ k = 1 m w k g k , G
=
\sum_{k=1}^{m}
w_k g_k, G = k = 1 ∑ m w k g k ,
也不能推出:
phenotype = f ( G ) \text{phenotype}
=
f(G) phenotype = f ( G )
只有一個維度。
因為:
不同 genes 可能作用於不同 pathways;
gene–environment interaction 存在;
pleiotropy 存在;
developmental stage 不同;
co-occurring conditions 不同。
所以:
continuous genetic liability ≠ unidimensional phenotype . \boxed{
\text{continuous genetic liability}
\neq
\text{unidimensional phenotype}.
} continuous genetic liability = unidimensional phenotype .
8. 2025 Twin Study:General Dimension 與 Secondary Dimensions 共存
Knyspel、Morneau-Vaillancourt 與 Eley 分析:
N = 10 , 454 N=10,454 N = 10 , 454
名 21 歲雙生子。
psychometric bifactor model 支持:
G ADHD G_{\text{ADHD}} G ADHD
這個 general symptom dimension,
同時保留:
I inattention , I_{\text{inattention}}, I inattention ,
H hyperactivity H_{\text{hyperactivity}} H hyperactivity
等 secondary dimensions。
因此:
general ADHD dimension + meaningful subdimensions \boxed{
\text{general ADHD dimension}
+
\text{meaningful subdimensions}
} general ADHD dimension + meaningful subdimensions
比單一 scalar 更合理。
遺傳與環境分解也沒有支持「所有 ADHD symptoms 共享一個完全同質遺傳因素」的簡單模型。
9. 從一維光譜升級為高維空間
本文因此不採:
x ∈ R x
\in
\mathbb R x ∈ R
作為完整 ADHD representation。
而採:
c = ( c 1 , c 2 , … , c d ) ∈ Ω C ⊆ R d . \boxed{
\mathbf c
=
(c_1,c_2,\ldots,c_d)
\in
\Omega_C
\subseteq
\mathbb R^d.
} c = ( c 1 , c 2 , … , c d ) ∈ Ω C ⊆ R d .
候選維度可包含:
c = ( A , I , W M , R , N , S , E , T , X , C context , … ) . \mathbf c
=
(
A,
I,
WM,
R,
N,
S,
E,
T,
X,
C_{\text{context}},
\ldots
). c = ( A , I , W M , R , N , S , E , T , X , C context , … ) .
例如:
attention allocation;
inhibitory control;
working memory;
reward sensitivity;
novelty sensitivity;
arousal regulation;
emotion regulation;
temporal organization;
switching/executive control;
context sensitivity。
這些不是正式臨床維度,只是中層候選坐標。
10. 配置空間不是診斷空間
本文定義:
Ω C = configuration space . \Omega_C
=
\text{configuration space}. Ω C = configuration space .
臨床觀察空間則為:
Ω Y = observable symptom/impairment space . \Omega_Y
=
\text{observable symptom/impairment space}. Ω Y = observable symptom / impairment space .
兩者間存在 mapping:
F : Ω C × Ω E → Ω Y . F:
\Omega_C
\times
\Omega_E
\rightarrow
\Omega_Y. F : Ω C × Ω E → Ω Y .
其中:
Ω E \Omega_E Ω E
包含 task、environment、development、support。
因此:
c ≠ y . \boxed{
\mathbf c
\neq
\mathbf y.
} c = y .
11. Clinical Diagnosis 是另一層算子
定義臨床判定:
D = Γ ( y , I , H , X ) . \mathcal D
=
\Gamma
\left(
\mathbf y,
I,
H,
X
\right). D = Γ ( y , I , H , X ) .
其中:
y \mathbf y y :症狀與行為表型;
I I I :功能損害;
H H H :發展史;
X X X :跨情境性、鑑別診斷與其他臨床資訊。
輸出:
D ∈ { 0 , 1 } . \mathcal D
\in
\{0,1\}. D ∈ { 0 , 1 } .
這裡的二元值只是抽象化「未符合/符合診斷」決策。
它不是對 DSM 或 ICD 的重建,也不是臨床計分器。
12. Continuous Input 可以產生 Categorical Output
即使:
y \mathbf y y
與:
I I I
是連續的,
Γ \Gamma Γ
仍可產生:
0 0 0
或:
1. 1. 1.
因此:
continuous phenotype → categorical clinical decision \boxed{
\text{continuous phenotype}
\rightarrow
\text{categorical clinical decision}
} continuous phenotype → categorical clinical decision
在數學上完全自然。
13. Clinical Threshold 是 Decision Threshold,不必是 Natural Cliff
若:
r ( y , I , H , X ) r(\mathbf y,I,H,X) r ( y , I , H , X )
為某種抽象 clinical-evidence score,
可概念化:
Γ = I [ r ≥ θ D ] . \Gamma
=
\mathbb I
\left[
r\geq\theta_D
\right]. Γ = I [ r ≥ θ D ] .
這不表示:
θ D \theta_D θ D
在自然界中必然對應一個:
biological discontinuity . \text{biological discontinuity}. biological discontinuity .
其功能可以是:
提高診斷一致性;
決定醫療服務入口;
支持風險/效益判斷;
建立可溝通類別。
14. 診斷類別可以有效,即使邊界不是自然斷崖
很多實際系統都使用 decision threshold。
例如:
continuous measurement → action category . \text{continuous measurement}
\rightarrow
\text{action category}. continuous measurement → action category .
因此:
operational threshold ≠ ontological cliff . \boxed{
\text{operational threshold}
\neq
\text{ontological cliff}.
} operational threshold = ontological cliff .
這是本文對 category–dimension 爭論最重要的修正。
15. Subthreshold 的最小定義
本文使用:
subthreshold \text{subthreshold} subthreshold
只表示:
存在部分 ADHD-related traits/difficulties,但目前不符合完整正式 ADHD 診斷要求。
抽象表示:
Γ ( ⋅ ) = 0 \Gamma(\cdot)=0 Γ ( ⋅ ) = 0
但:
∥ y ADHD-related ∥ > 0. \|\mathbf y_{\text{ADHD-related}}\|>0. ∥ y ADHD-related ∥ > 0.
因此:
subthreshold ≠ undiagnosed clinical ADHD by definition . \boxed{
\text{subthreshold}
\neq
\text{undiagnosed clinical ADHD by definition}.
} subthreshold = undiagnosed clinical ADHD by definition .
16. Subthreshold 也可能有真正功能困難
可以存在:
Γ = 0 , \Gamma=0, Γ = 0 ,
但:
I > 0. I>0. I > 0.
此時:
support need ≠ diagnostic status . \boxed{
\text{support need}
\neq
\text{diagnostic status}.
} support need = diagnostic status .
一個人可能需要:
organization support;
sleep intervention;
educational accommodation;
environmental restructuring;
psychological treatment for another condition;
而不代表 ADHD diagnosis 必然成立。
17. 2025 Subthreshold Review 的啟示與限制
2025 年 Ogundele 等人的 narrative review 討論 subthreshold autism/ADHD。
作者主張,若未達正式 NDD criteria,但存在顯著、持續的 impairment,可以考慮記錄 subthreshold condition。
這是一項臨床討論與作者建議。
它不是:
DSM/ICD 已新增正式 subthreshold ADHD diagnosis . \boxed{
\text{DSM/ICD 已新增正式 subthreshold ADHD diagnosis}.
} DSM / ICD 已新增正式 subthreshold ADHD diagnosis .
因此本文只採用:
clinically relevant below-threshold difficulties \text{clinically relevant below-threshold difficulties} clinically relevant below-threshold difficulties
這個較弱概念。
18. 不要把「接近診斷區」叫做「其實已經有」
若個體:
c i \mathbf c_i c i
與某臨床群體平均 configuration:
μ ADHD \boldsymbol\mu_{\text{ADHD}} μ ADHD
很接近,
也不能推出:
D i = 1. \mathcal D_i=1. D i = 1.
所以:
d ( c i , μ ADHD ) ↓ ⇏ ADHD diagnosis . \boxed{
d
\left(
\mathbf c_i,
\boldsymbol\mu_{\text{ADHD}}
\right)
\downarrow
\not\Rightarrow
\text{ADHD diagnosis}.
} d ( c i , μ ADHD ) ↓ ⇒ ADHD diagnosis .
19. Configuration Distance
若有標準化 configuration coordinates,可定義:
d C ( c i , c j ) . d_C
\left(
\mathbf c_i,\mathbf c_j
\right). d C ( c i , c j ) .
例如 Mahalanobis distance:
d M = ( c i − μ ) ⊤ Σ − 1 ( c i − μ ) . d_M
=
\sqrt{
(\mathbf c_i-\boldsymbol\mu)^{\top}
\Sigma^{-1}
(\mathbf c_i-\boldsymbol\mu)
}. d M = ( c i − μ ) ⊤ Σ − 1 ( c i − μ ) .
但:
d M d_M d M
只能表示多維統計距離。
它不是:
diagnostic probability . \text{diagnostic probability}. diagnostic probability .
20. 相似配置可以有不同診斷結果
若:
c i ≈ c j , \mathbf c_i
\approx
\mathbf c_j, c i ≈ c j ,
但:
I i ≠ I j , I_i\neq I_j, I i = I j ,
或:
H i ≠ H j , H_i\neq H_j, H i = H j ,
或:
X i ≠ X j , X_i\neq X_j, X i = X j ,
可以:
Γ i ≠ Γ j . \Gamma_i\neq\Gamma_j. Γ i = Γ j .
因此:
configuration similarity ≠ clinical equivalence . \boxed{
\text{configuration similarity}
\neq
\text{clinical equivalence}.
} configuration similarity = clinical equivalence .
21. 不同配置也可以產生相似表型
反過來:
c i ≠ c j \mathbf c_i
\neq
\mathbf c_j c i = c j
可能仍有:
y i ≈ y j . \mathbf y_i
\approx
\mathbf y_j. y i ≈ y j .
例如 attention failure 可以來自:
arousal instability;
distractor competition;
working-memory failure;
sleep-like lapse;
high switching;
motivational mismatch。
因此:
same phenotype ← multiple configurations . \boxed{
\text{same phenotype}
\leftarrow
\text{multiple configurations}.
} same phenotype ← multiple configurations .
22. Many-to-One Mapping
形式上:
F : Ω C → Ω Y F:
\Omega_C
\rightarrow
\Omega_Y F : Ω C → Ω Y
可能是 many-to-one。
即:
F ( c 1 ) = F ( c 2 ) F(\mathbf c_1)
=
F(\mathbf c_2) F ( c 1 ) = F ( c 2 )
而:
c 1 ≠ c 2 . \mathbf c_1\neq\mathbf c_2. c 1 = c 2 .
這是 ADHD heterogeneity 的一個自然數學表示。
23. One-to-Many Across Contexts
同一配置:
c \mathbf c c
在不同環境:
e 1 , e 2 e_1,e_2 e 1 , e 2
可產生:
F ( c , e 1 ) ≠ F ( c , e 2 ) . F(\mathbf c,e_1)
\neq
F(\mathbf c,e_2). F ( c , e 1 ) = F ( c , e 2 ) .
因此:
same person + different context → different phenotype . \boxed{
\text{same person}
+
\text{different context}
\rightarrow
\text{different phenotype}.
} same person + different context → different phenotype .
這與前七篇的 dynamic-configuration model 相容。
24. 密度,而不是天然牆壁
令一般人口 configuration density:
ρ ( c ) . \rho(\mathbf c). ρ ( c ) .
ADHD-related diagnosed sample density:
ρ D ( c ) . \rho_D(\mathbf c). ρ D ( c ) .
可能:
ρ D \rho_D ρ D
在某些空間區域較高,
但未必存在:
∂ Ω D \partial\Omega_D ∂ Ω D
這種完全沒有重疊的天然硬牆。
因此:
density enrichment ≠ perfect separability . \boxed{
\text{density enrichment}
\neq
\text{perfect separability}.
} density enrichment = perfect separability .
25. Local Clusters 可以存在於 Continuous Space
假設:
ρ ( c ) \rho(\mathbf c) ρ ( c )
具有多個局部峰:
μ 1 , μ 2 , … , μ K . \boldsymbol\mu_1,
\boldsymbol\mu_2,\ldots,\boldsymbol\mu_K. μ 1 , μ 2 , … , μ K .
則:
continuum + clusters \boxed{
\text{continuum}
+
\text{clusters}
} continuum + clusters
可以同時成立。
這就是 2026 JAMA Psychiatry biotype study 對本篇最重要的啟示。
26. 2026 Biotypes:不是「三種真正 ADHD」的最終答案
Pan 等人在 pediatric datasets 中使用 normative modeling 與 semisupervised clustering。
discovery cohort:
446 446 446
名 ADHD 兒童與:
708 708 708
名 controls。
external validation cohort 包含:
554 554 554
名 ADHD cases。
得到三個 biotypes:
severe-combined+emotional dysregulation;
predominantly hyperactive/impulsive;
predominantly inattentive。
研究結論明確稱其提供:
dimensional and categorical insights . \boxed{
\text{dimensional and categorical insights}.
} dimensional and categorical insights .
但本文不把三個 biotypes 視為正式新診斷。
27. Biotype 不等於 Clinical Subtype
目前影像 biotype:
B k neuro B_k^{\text{neuro}} B k neuro
與臨床 subtype:
B k clinical B_k^{\text{clinical}} B k clinical
不能直接等同。
因此:
data-driven biotype ≠ validated clinical diagnostic type . \boxed{
\text{data-driven biotype}
\neq
\text{validated clinical diagnostic type}.
} data-driven biotype = validated clinical diagnostic type .
仍需:
replication;
individual-level reliability;
treatment prediction;
longitudinal stability;
cost-benefit evaluation。
28. Normative Modeling 的重要觀念
Pan 等人的方法先建立:
normative distribution \text{normative distribution} normative distribution
再計算個體 deviation:
z i = x i − μ norm σ norm . z_i
=
\frac{
x_i-\mu_{\text{norm}}
}{
\sigma_{\text{norm}}
}. z i = σ norm x i − μ norm .
這比單純:
ADHD mean − control mean \text{ADHD mean}
-
\text{control mean} ADHD mean − control mean
更接近個體化。
因此本篇引入:
individual deviation profile . \boxed{
\text{individual deviation profile}.
} individual deviation profile .
29. 配置偏差向量
定義:
z i = ( z i 1 , z i 2 , … , z i d ) . \mathbf z_i
=
\left(
z_{i1},
z_{i2},
\ldots,z_{id}
\right). z i = ( z i 1 , z i 2 , … , z i d ) .
每一維是相對 reference population 的 deviation。
但:
z i \mathbf z_i z i
不代表 pathology map。
它只是:
relative position . \text{relative position}. relative position .
30. 正常範圍本身不是單一點
reference population 不是:
c = 0. \mathbf c=\mathbf0. c = 0 .
而是分布:
p ref ( c ) . p_{\text{ref}}(\mathbf c). p ref ( c ) .
所以:
normal variation ≠ zero variation . \boxed{
\text{normal variation}
\neq
\text{zero variation}.
} normal variation = zero variation .
31. 診斷不能由「偏離平均」直接得到
一個人可能:
∣ z k ∣ ≫ 0 |z_k|\gg0 ∣ z k ∣ ≫ 0
但沒有 impairment。
另一人:
∣ z k ∣ |z_k| ∣ z k ∣
不極端,
卻在特定生活需求下有嚴重功能困難。
因此:
statistical atypicality ≠ clinical disorder . \boxed{
\text{statistical atypicality}
\neq
\text{clinical disorder}.
} statistical atypicality = clinical disorder .
32. 遺傳異質性支持「一個 ADHD 裡有很多路徑」
2024 Nature Genetics 對:
14 , 084 14,084 14 , 084
名 diagnosed ADHD individuals 進行 case-only genetic heterogeneity research。
研究顯示不同 ADHD-adjacent profiles,例如:
ADHD+ASD;
ADHD+substance use disorder;
adulthood-first-diagnosed ADHD;
具有可區分的 polygenic score patterns。
因此:
same diagnostic label ≠ same polygenic profile . \boxed{
\text{same diagnostic label}
\neq
\text{same polygenic profile}.
} same diagnostic label = same polygenic profile .
33. 遺傳異質性不等於可以基因診斷個人
即使群體層:
P G S A ≠ P G S B , PGS_A\neq PGS_B, P G S A = P G S B ,
也不能推出:
individual diagnosis = f ( P G S ) . \text{individual diagnosis}
=
f(PGS). individual diagnosis = f ( P GS ) .
現階段:
polygenic association ≠ clinical diagnostic test . \boxed{
\text{polygenic association}
\neq
\text{clinical diagnostic test}.
} polygenic association = clinical diagnostic test .
34. 高維配置空間的多尺度結構
本文將 ADHD-related space 暫分四層:
34.1 Trait Layer
Ω T . \Omega_T. Ω T .
連續 symptoms/traits。
34.2 Cognitive Configuration Layer
Ω C . \Omega_C. Ω C .
認知與狀態維度。
34.3 Functional Layer
Ω I . \Omega_I. Ω I .
日常功能損害。
34.4 Clinical Decision Layer
Ω D . \Omega_D. Ω D .
正式診斷與醫療決策。
因此:
Ω T ≠ Ω C ≠ Ω I ≠ Ω D . \boxed{
\Omega_T
\neq
\Omega_C
\neq
\Omega_I
\neq
\Omega_D.
} Ω T = Ω C = Ω I = Ω D .
35. 不能用一條 Axis 代替四層
錯誤模型:
x = ADHD amount . x
=
\text{ADHD amount}. x = ADHD amount .
更合理:
X = ( t , c , i , h , e ) . \mathbf X
=
\left(
\mathbf t,
\mathbf c,
\mathbf i,
\mathbf h,
\mathbf e
\right). X = ( t , c , i , h , e ) .
其中:
t \mathbf t t :traits;
c \mathbf c c :cognitive configuration;
i \mathbf i i :impairment profile;
h \mathbf h h :developmental history;
e \mathbf e e :environment/context。
36. Functional Impairment 是向量
成人功能至少可以包含:
i = ( I academic , I work , I financial , I relationship , I daily , I safety ) . \mathbf i
=
\left(
I_{\text{academic}},
I_{\text{work}},
I_{\text{financial}},
I_{\text{relationship}},
I_{\text{daily}},
I_{\text{safety}}
\right). i = ( I academic , I work , I financial , I relationship , I daily , I safety ) .
因此:
I I I
不必只有一個總分。
37. 不同人可以在不同功能域跨門檻
例如:
I work > θ , I_{\text{work}}>\theta, I work > θ ,
但:
I relationship < θ . I_{\text{relationship}}<\theta. I relationship < θ .
另一人相反。
所以:
functional impairment = profile , \boxed{
\text{functional impairment}
=
\text{profile},
} functional impairment = profile ,
而不是純 scalar。
38. Impairment Surface
給定 configuration:
c \mathbf c c
與 context:
e , \mathbf e, e ,
定義:
I = Φ ( c , e ) . I
=
\Phi
\left(
\mathbf c,\mathbf e
\right). I = Φ ( c , e ) .
這形成:
impairment surface . \boxed{
\text{impairment surface}.
} impairment surface .
同一 configuration 在不同 context 對應不同高度。
39. Clinical Region 不是單純球體
若將 formal diagnosis region 抽象表示為:
R D ⊂ Ω T × Ω I × Ω H , \mathcal R_D
\subset
\Omega_T
\times
\Omega_I
\times
\Omega_H, R D ⊂ Ω T × Ω I × Ω H ,
它不應被想像成:
∥ c − μ ∥ < r \|\mathbf c-\boldsymbol\mu\|<r ∥ c − μ ∥ < r
的簡單球體。
臨床規則通常包含:
symptom pattern;
impairment;
duration;
developmental onset;
cross-setting evidence;
differential diagnosis。
所以:
R D is a decision region, not a Euclidean cluster . \boxed{
\mathcal R_D
\text{ is a decision region, not a Euclidean cluster}.
} R D is a decision region, not a Euclidean cluster .
40. Subthreshold Region
可概念化:
R sub = { x : Γ ( x ) = 0 ∧ ∥ y ADHD-related ∥ > 0 } . \mathcal R_{\text{sub}}
=
\left\{
x:
\Gamma(x)=0
\land
\|\mathbf y_{\text{ADHD-related}}\|>0
\right\}. R sub = { x : Γ ( x ) = 0 ∧ ∥ y ADHD-related ∥ > 0 } .
但:
R sub \mathcal R_{\text{sub}} R sub
內部同樣高度異質。
可能包括:
transient traits;
mild persistent traits;
impairment from another condition;
compensated profiles;
developmental variation;
measurement noise。
因此:
subthreshold ≠ one latent disorder . \boxed{
\text{subthreshold}
\neq
\text{one latent disorder}.
} subthreshold = one latent disorder .
41. Near-Threshold Instability
若某人:
r ≈ θ D , r\approx\theta_D, r ≈ θ D ,
小幅 measurement/context change 可造成:
Γ t = 0 \Gamma_t=0 Γ t = 0
與:
Γ t + 1 = 1. \Gamma_{t+1}=1. Γ t + 1 = 1.
這不必表示:
biology suddenly changed . \text{biology suddenly changed}. biology suddenly changed .
而可能只是:
classification near a decision boundary is sensitive to context and measurement . \boxed{
\text{classification near a decision boundary is sensitive to context and measurement}.
} classification near a decision boundary is sensitive to context and measurement .
42. 診斷變化不等於身份翻轉
一個人某時:
Γ t = 1 \Gamma_t=1 Γ t = 1
後來:
Γ t + 1 = 0 \Gamma_{t+1}=0 Γ t + 1 = 0
可能反映:
symptom remission;
developmental change;
treatment;
context;
measurement;
criterion differences。
因此:
diagnostic status over time ≠ immutable cognitive identity . \boxed{
\text{diagnostic status over time}
\neq
\text{immutable cognitive identity}.
} diagnostic status over time = immutable cognitive identity .
43. Adult Review 的重要提醒:即使不再符合診斷,困難仍可能存在
2025 World Psychiatry 成人 ADHD 綜述指出,部分 childhood-onset ADHD individuals 到成年後可能仍有 impairing symptoms,即使不再滿足完整 formal diagnostic criteria。
這再次支持:
diagnostic threshold crossing ≠ all-or-none disappearance of traits . \boxed{
\text{diagnostic threshold crossing}
\neq
\text{all-or-none disappearance of traits}.
} diagnostic threshold crossing = all-or-none disappearance of traits .
44. Persistent Trait Without Diagnosis
可能:
Γ t = 0 , \Gamma_t=0, Γ t = 0 ,
但:
y t ≠ 0. \mathbf y_t\neq\mathbf0. y t = 0 .
這並不矛盾。
類別判定變化可以與連續 trait persistence 同時存在。
45. Transdiagnostic Overlap
ADHD-related dimensions 與:
autism;
anxiety;
depression;
sleep disorders;
learning disorders;
substance-related conditions;
可能共享部分:
Ω C \Omega_C Ω C
區域。
因此:
shared dimension ≠ same disorder . \boxed{
\text{shared dimension}
\neq
\text{same disorder}.
} shared dimension = same disorder .
46. 診斷仍需要 Differential Information
若兩種 disorder:
D 1 , D 2 D_1,
D_2 D 1 , D 2
在某些 configuration 維度重疊,
則需要:
X X X
中的其他 evidence 進行區分。
所以高維 dimensional model 不是:
diagnosis-free model . \text{diagnosis-free model}. diagnosis-free model .
反而更需要嚴格 differential diagnosis。
47. Pure Spectrum Model 也會失敗
若只寫:
everyone is a little ADHD , \text{everyone is a little ADHD}, everyone is a little ADHD ,
會產生嚴重概念錯誤。
因為:
trait presence \text{trait presence} trait presence
與:
persistent clinically significant disorder \text{persistent clinically significant disorder} persistent clinically significant disorder
不是同一件事。
因此本文明確拒絕:
everyone is on the ADHD spectrum ⇒ everyone has ADHD . \boxed{
\text{everyone is on the ADHD spectrum}
\Rightarrow
\text{everyone has ADHD}.
} everyone is on the ADHD spectrum ⇒ everyone has ADHD .
48. Pure Binary Model 也會失敗
反過來若只寫:
A D H D ∈ { 0 , 1 } ADHD\in\{0,1\} A D H D ∈ { 0 , 1 }
並假設兩群所有底層特徵完全分離,
也無法容納:
symptom continuum;
subthreshold impairment;
remission;
context dependence;
genetic continuity;
within-diagnosis heterogeneity。
所以:
binary decision ≠ binary ontology . \boxed{
\text{binary decision}
\neq
\text{binary ontology}.
} binary decision = binary ontology .
49. Continuous-Plus-Cluster Model
本文提出:
Ω C = continuous space with possible local density structure . \boxed{
\Omega_C
=
\text{continuous space with possible local density structure}.
} Ω C = continuous space with possible local density structure .
概念上可以用 mixture density:
p ( c ) = ∑ k = 1 K π k p k ( c ) . p(\mathbf c)
=
\sum_{k=1}^{K}
\pi_k
p_k(\mathbf c). p ( c ) = k = 1 ∑ K π k p k ( c ) .
但:
K K K
不預設為 3。
2026 biotype study 找到 3 群,不代表自然界永遠只有 3 群。
50. Cluster Stability
若存在 cluster:
C k , C_k, C k ,
需要檢驗:
Stability ( C k ) \operatorname{Stability}(C_k) Stability ( C k )
跨:
cohort;
age;
culture;
sex;
medication;
scanner/instrument;
time。
如果 cluster 不穩定,就不能臨床本體化。
51. Cluster Membership 也可以是機率
不必:
x ∈ C 1 x\in C_1 x ∈ C 1
或:
x ∈ C 2 . x\in C_2. x ∈ C 2 .
可以:
P ( C k ∣ x ) . P(C_k\mid x). P ( C k ∣ x ) .
因此:
fuzzy membership \boxed{
\text{fuzzy membership}
} fuzzy membership
可能比硬分類更符合異質資料。
52. Category、Dimension 與 Cluster 的三層統一
本文最終統一:
Dimension → describes variation \boxed{
\text{Dimension}
\rightarrow
\text{describes variation}
} Dimension → describes variation
Cluster → describes local structure \boxed{
\text{Cluster}
\rightarrow
\text{describes local structure}
} Cluster → describes local structure
Category → supports decisions \boxed{
\text{Category}
\rightarrow
\text{supports decisions}
} Category → supports decisions
三者不是競爭對手。
53. Clinical Diagnosis 的價值不依賴於「自然斷點」
診斷可以提供:
treatment access;
communication;
prognosis;
support eligibility;
research inclusion;
risk management。
所以:
clinical usefulness ≠ proof of natural-kind discreteness . \boxed{
\text{clinical usefulness}
\neq
\text{proof of natural-kind discreteness}.
} clinical usefulness = proof of natural-kind discreteness .
54. 反過來,Dimension 也不能自行決定 Treatment
知道:
c i \mathbf c_i c i
位於某個連續位置,
並不能直接推出:
treatment i . \text{treatment}_i. treatment i .
治療還取決於:
impairment;
preference;
comorbidity;
risk;
evidence base;
contraindications;
goals。
所以:
dimensional description ≠ clinical prescription . \boxed{
\text{dimensional description}
\neq
\text{clinical prescription}.
} dimensional description = clinical prescription .
55. CCSH 十二項核心命題
CC-H1:連續性命題
至少部分 ADHD-related traits:
T k T_k T k
在一般人口中近似連續分布。
CC-H2:診斷非等同性命題
T k > 0 ⇏ D = 1. T_k>0
\not\Rightarrow
\mathcal D=1. T k > 0 ⇒ D = 1.
CC-H3:症狀—損害分離命題
S symptom ≠ I impairment . S_{\text{symptom}}
\neq
I_{\text{impairment}}. S symptom = I impairment .
CC-H4:高維命題
ADHD-related variation 不能被單一 scalar 完整表示。
CC-H5:Continuous-Plus-Cluster 命題
continuous variation + local clustering \text{continuous variation}
+
\text{local clustering} continuous variation + local clustering
可以同時存在。
CC-H6:Many-to-One 命題
不同 configurations 可產生相似 phenotype。
CC-H7:Contextual One-to-Many 命題
同一 configuration 在不同 context 可產生不同 phenotype。
CC-H8:Subthreshold Non-Identity 命題
subthreshold traits ≠ hidden clinical ADHD by definition . \text{subthreshold traits}
\neq
\text{hidden clinical ADHD by definition}. subthreshold traits = hidden clinical ADHD by definition .
CC-H9:Decision-Boundary 命題
正式診斷 boundary 可以具有臨床效用,而不要求 underlying natural cliff。
CC-H10:Fuzzy-Biotype 命題
biotype membership 若存在,可能是 probabilistic 而非硬分割。
CC-H11:Transdiagnostic Overlap 命題
部分 cognitive dimensions 可以跨診斷共享,但共享不取消 differential diagnosis。
CC-H12:增量價值命題
若高維 continuous-plus-cluster model 在控制 symptom score 與現行 diagnosis 後,不能提高 impairment/course/treatment-outcome prediction:
Δ R 2 ≈ 0 , \Delta R^2\approx0, Δ R 2 ≈ 0 ,
則 CCSH 應被簡化。
56. 實驗一:Population Density Mapping
在大型一般人口樣本測:
c i . \mathbf c_i. c i .
估計:
p ( c ) . p(\mathbf c). p ( c ) .
檢查:
unimodal;
multimodal;
heavy-tail;
manifold;
local clusters。
不預先切 ADHD/control。
57. 實驗二:Symptom–Impairment Surface
同時測:
y i \mathbf y_i y i
與:
i i . \mathbf i_i. i i .
建模:
i = F ( y , e ) . \mathbf i
=
F(\mathbf y,\mathbf e). i = F ( y , e ) .
檢驗:
∂ I ∂ S \frac{
\partial I
}{
\partial S
} ∂ S ∂ I
是否在高 symptom 區突然改變。
58. 實驗三:Threshold Robustness
對同一 dataset 改變:
θ D . \theta_D. θ D .
觀察:
classification stability;
impairment prediction;
treatment response;
false positive/negative trade-off。
如果小幅 threshold change 造成 outcome validity 崩潰,現行 boundary 需要更多研究。
59. 實驗四:Normative Configuration Modeling
建立:
p ref ( c ∣ a g e , s e x , … ) . p_{\text{ref}}(\mathbf c\mid age,sex,\ldots). p ref ( c ∣ a g e , se x , … ) .
計算 individual deviation:
z i . \mathbf z_i. z i .
再測:
z i → I i . \mathbf z_i
\rightarrow
I_i. z i → I i .
不直接用 z \mathbf z z 做 diagnosis。
60. 實驗五:Cluster Replication
在 discovery cohort 建:
C 1 , … , C K . C_1,\ldots,C_K. C 1 , … , C K .
在 external cohort 檢驗:
ARI , \operatorname{ARI}, ARI ,
NMI , \operatorname{NMI}, NMI ,
stability . \operatorname{stability}. stability .
若 cluster 不可外部重現,不應稱 biotype。
61. 實驗六:Longitudinal Boundary Crossing
追蹤:
c t , y t , I t , Γ t . \mathbf c_t,
\mathbf y_t,I_t,\Gamma_t. c t , y t , I t , Γ t .
檢查:
Γ t : 0 → 1 \Gamma_t:
0\rightarrow1 Γ t : 0 → 1
時,底層 configuration 是否真的出現非連續改變。
若沒有:
diagnostic transition ≠ configuration phase transition . \boxed{
\text{diagnostic transition}
\neq
\text{configuration phase transition}.
} diagnostic transition = configuration phase transition .
62. 實驗七:Subthreshold Outcome Study
建立:
low-trait controls;
subthreshold traits without impairment;
subthreshold traits with impairment;
clinical ADHD。
比較:
course;
function;
help-seeking;
comorbidity;
intervention needs。
這可以真正回答:
subthreshold \text{subthreshold} subthreshold
是否具有獨立臨床意義。
63. 實驗八:Cross-Diagnostic Configuration Mapping
同時納入:
ADHD;
autism;
anxiety;
depression;
sleep disorders;
controls。
建立:
Ω C . \Omega_C. Ω C .
檢驗:
shared dimensions \text{shared dimensions} shared dimensions
與:
diagnosis-specific combinations . \text{diagnosis-specific combinations}. diagnosis-specific combinations .
64. 實驗九:Treatment-Outcome Geometry
測:
P ( R treat ∣ c ) . P
\left(
R_{\text{treat}}
\mid
\mathbf c
\right). P ( R treat ∣ c ) .
若 configuration space 能預測 treatment response,而 diagnosis alone 不能,才具有 precision-medicine 潛力。
在此之前:
configuration model ≠ treatment selector . \boxed{
\text{configuration model}
\neq
\text{treatment selector}.
} configuration model = treatment selector .
65. 模型失敗條件
CCSH 應被削弱,如果:
ADHD-related trait distributions 反覆呈現清晰自然雙峰;
symptom–impairment relation 在獨立樣本中存在穩定自然斷點;
單一 scalar model 已足以解釋 cognition、impairment 與 course;
clusters 無法外部重現;
high-dimensional model 嚴重 overfit;
subthreshold group 與 low-trait controls 在所有 clinically relevant outcomes 都無差異;
configuration variables 無法提高 treatment/course prediction;
dimensional measurements 無法跨文化或跨年齡穩定操作化。
若:
P CCSH,out ≤ P binary,out , P_{\text{CCSH,out}}
\leq
P_{\text{binary,out}}, P CCSH,out ≤ P binary,out ,
且:
P CCSH,out ≤ P single-axis,out , P_{\text{CCSH,out}}
\leq
P_{\text{single-axis,out}}, P CCSH,out ≤ P single-axis,out ,
則高維模型沒有保留必要。
66. 最重要反例一:Formal ADHD Diagnosis 仍有 Validity
2024 Nature Reviews Disease Primers 等現代綜述仍強調:
standard diagnostic criteria can identify a reliable and clinically meaningful ADHD syndrome . \boxed{
\text{standard diagnostic criteria can identify
a reliable and clinically meaningful ADHD syndrome}.
} standard diagnostic criteria can identify a reliable and clinically meaningful ADHD syndrome .
所以本文不是:
ADHD diagnosis is arbitrary and meaningless . \text{ADHD diagnosis is arbitrary and meaningless}. ADHD diagnosis is arbitrary and meaningless .
更精確:
the underlying liability may be continuous while the diagnostic construct remains clinically valid . \boxed{
\text{the underlying liability may be continuous
while the diagnostic construct remains clinically valid}.
} the underlying liability may be continuous while the diagnostic construct remains clinically valid .
67. 最重要反例二:不是所有 Continuum 都必須是一維
2025 adult twin data 正好顯示:
general dimension + secondary dimensions . \text{general dimension}
+
\text{secondary dimensions}. general dimension + secondary dimensions .
因此本文拒絕:
ADHD spectrum = one straight line . \boxed{
\text{ADHD spectrum}
=
\text{one straight line}.
} ADHD spectrum = one straight line .
68. 最重要反例三:Biotype Evidence 不等於終結 Spectrum
2026 JAMA study 的 clustering 建立在:
normative dimensional deviations \text{normative dimensional deviations} normative dimensional deviations
之上。
所以其方法本身說明:
clusters can emerge inside a dimensional framework . \boxed{
\text{clusters can emerge inside a dimensional framework}.
} clusters can emerge inside a dimensional framework .
這不是 category 對 dimension 的勝利。
69. 最重要反例四:Subthreshold 不是偷渡診斷
即使:
I > 0 I>0 I > 0
且:
Γ = 0 , \Gamma=0, Γ = 0 ,
也不能由本文推出:
Γ 應改成 1. \Gamma\text{ 應改成 }1. Γ 應改成 1.
臨床可能需要:
重新評估;
其他診斷;
非診斷型支持;
追蹤;
環境調整。
所以:
need for help ≠ need for a specific diagnosis . \boxed{
\text{need for help}
\neq
\text{need for a specific diagnosis}.
} need for help = need for a specific diagnosis .
70. 與前七篇整合
第 1 篇提出:
C t = dynamic configuration . \mathbf C_t
=
\text{dynamic configuration}. C t = dynamic configuration .
第 2–6 篇逐層展開:
N t , Z t , Π t , X t , G t , Q t , P ^ t , P t . \mathbf N_t,
\mathbf Z_t,
\Pi_t,
\mathbf X_t,
\mathcal G_t,
Q_t,
\widehat P_t,
P_t. N t , Z t , Π t , X t , G t , Q t , P t , P t .
第 7 篇加入生命史:
D t , S t , K t , V t . D_t,S_t,K_t,V_t. D t , S t , K t , V t .
本篇現在把所有個體狀態嵌入:
Ω C ⊆ R d . \boxed{
\Omega_C
\subseteq
\mathbb R^d.
} Ω C ⊆ R d .
並將臨床 diagnosis 明確放在另一層:
D = Γ ( y , I , H , X ) . \boxed{
\mathcal D
=
\Gamma
\left(
\mathbf y,
I,
H,
X
\right).
} D = Γ ( y , I , H , X ) .
因此整套理論第一次明確區分:
ontology of variation \boxed{
\text{ontology of variation}
} ontology of variation
與:
clinical decision rule . \boxed{
\text{clinical decision rule}.
} clinical decision rule .
71. 系列目前的統一表示
底層狀態:
c i , t ∈ Ω C . \mathbf c_{i,t}
\in
\Omega_C. c i , t ∈ Ω C .
環境耦合:
y i , t = F ( c i , t , e i , t ) . \mathbf y_{i,t}
=
F
\left(
\mathbf c_{i,t},
\mathbf e_{i,t}
\right). y i , t = F ( c i , t , e i , t ) .
功能損害:
i i , t = Φ ( c i , t , e i , t , d i , t ) . \mathbf i_{i,t}
=
\Phi
\left(
\mathbf c_{i,t},
\mathbf e_{i,t},
\mathbf d_{i,t}
\right). i i , t = Φ ( c i , t , e i , t , d i , t ) .
臨床判定:
D i , t = Γ ( y i , t , i i , t , H i , X i , t ) . \boxed{
\mathcal D_{i,t}
=
\Gamma
\left(
\mathbf y_{i,t},
\mathbf i_{i,t},
H_i,
X_{i,t}
\right).
} D i , t = Γ ( y i , t , i i , t , H i , X i , t ) .
所以:
c continuous ∧ D categorical \boxed{
\mathbf c
\text{ continuous}
\quad\land\quad
\mathcal D
\text{ categorical}
} c continuous ∧ D categorical
完全可以同時成立。
72. 本文不主張的內容
本文不主張:
ADHD diagnosis 沒有意義;
ADHD 不是疾病/障礙;
診斷 threshold 可以任意取消;
everyone has ADHD;
every ADHD-like trait is pathological;
subthreshold 等於 hidden ADHD;
near-threshold 等於一定應被診斷;
functional impairment 可以忽略;
genetics 已證明 ADHD 只有單一 continuum;
2026 biotypes 是 ADHD 最終三分類;
brain imaging 已可進行個人 ADHD diagnosis;
polygenic scores 已可用於個人 diagnosis;
high-dimensional configuration 可取代 DSM/ICD;
診斷 category 與 dimension 必須二選一;
shared transdiagnostic traits 代表不同 disorders 都是一樣;
statistical atypicality 等於 clinical disorder;
clinical diagnosis 等於固定終身身份;
本模型可用來替任何人判斷「接近 ADHD 幾成」。
73. 結論
「ADHD 是類別還是光譜?」若被理解成二選一,已經不足以描述現有證據。
較完整的模型是:
continuous high-dimensional variation + possible local clusters + categorical clinical decisions . \boxed{
\text{continuous high-dimensional variation}
+
\text{possible local clusters}
+
\text{categorical clinical decisions}.
} continuous high-dimensional variation + possible local clusters + categorical clinical decisions .
2024–2025 的 population、QoL、twin 與 GWAS evidence 支持 ADHD-related symptoms 與 liability 的連續性;2026 的 biotype research 又顯示連續 normative deviations 中可以出現可重現局部結構。
因此 clinical ADHD 可以被概念化為:
a clinically meaningful decision region constructed over continuous and heterogeneous variation , \boxed{
\text{a clinically meaningful decision region
constructed over continuous and heterogeneous variation},
} a clinically meaningful decision region constructed over continuous and heterogeneous variation ,
而不是:
a point where human cognition suddenly changes species . \text{a point where human cognition suddenly changes species}. a point where human cognition suddenly changes species .
但這個說法不削弱 diagnosis。
它反而要求更加精確地區分:
trait , \text{trait}, trait ,
configuration , \text{configuration}, configuration ,
impairment , \text{impairment}, impairment ,
developmental history , \text{developmental history}, developmental history ,
diagnosis . \text{diagnosis}. diagnosis .
尤其:
subthreshold traits ≠ clinical ADHD . \boxed{
\text{subthreshold traits}
\neq
\text{clinical ADHD}.
} subthreshold traits = clinical ADHD .
一個人可以具有 ADHD-related traits 而不符合 ADHD;也可以在未達 ADHD 診斷時具有真正需要處理的功能困難。
本文最終提出:
Ω C = continuous configuration space , \boxed{
\Omega_C
=
\text{continuous configuration space},
} Ω C = continuous configuration space ,
ρ ( c ) = population density with possible local structure , \boxed{
\rho(\mathbf c)
=
\text{population density with possible local structure},
} ρ ( c ) = population density with possible local structure ,
以及:
D = Γ ( symptoms , impairment , history , context , differential evidence ) . \boxed{
\mathcal D
=
\Gamma
\left(
\text{symptoms},
\text{impairment},
\text{history},
\text{context},
\text{differential evidence}
\right).
} D = Γ ( symptoms , impairment , history , context , differential evidence ) .
最終可證偽問題是:
Does a high-dimensional continuous-plus-cluster model predict impairment, course, and treatment-relevant outcomes better than either a pure binary model or a pure single-axis spectrum? \boxed{
\text{Does a high-dimensional continuous-plus-cluster model
predict impairment, course, and treatment-relevant outcomes
better than either a pure binary model
or a pure single-axis spectrum?}
} Does a high-dimensional continuous-plus-cluster model predict impairment, course, and treatment-relevant outcomes better than either a pure binary model or a pure single-axis spectrum?
如果不能,CCSH 應被簡化。
如果可以,ADHD 就可以同時保持臨床診斷的操作價值,又允許底層研究從二元標籤進入更精細的配置幾何。
參考文獻
Arildskov, T. W., Thomsen, P. H., Sonuga-Barke, E. J. S., Lambek, R., Østergaard, S. D., & Virring, A. Is Attention-Deficit/Hyperactivity Disorder (ADHD) a Dimension or a Category? What Does the Relationship Between ADHD Traits and Psychosocial Quality of Life Tell Us? Journal of Attention Disorders . 2024;28(7):1035–1044. DOI: 10.1177/10870547231222228.
Arildskov, T. W., Sonuga-Barke, E. J. S., Thomsen, P. H., Virring, A., & Østergaard, S. D. How much impairment is required for ADHD? No evidence of a discrete threshold. Journal of Child Psychology and Psychiatry . 2022;63(2):229–237. DOI: 10.1111/jcpp.13440.
van der Laan, C. M., Ip, H. F., Schipper, M., et al. Genome-wide association meta-analysis of childhood ADHD symptoms and diagnosis identifies new loci and potential effector genes. Nature Genetics . 2025;57:2427–2435. DOI: 10.1038/s41588-025-02295-y.
Knyspel, J., Morneau-Vaillancourt, G., & Eley, T. C. Using Bifactor Twin Modeling to Assess the Genetic and Environmental Dimensionality of Adult ADHD Symptoms. Behavior Genetics . 2025;55:1–11. DOI: 10.1007/s10519-024-10204-y.
Pan, N., Long, Y., Qin, K., et al. Mapping ADHD Heterogeneity and Biotypes by Topological Deviations in Morphometric Similarity Networks. JAMA Psychiatry . 2026;83(5):478–490. DOI: 10.1001/jamapsychiatry.2026.0001.
Schork, A. J., et al. Polygenic profiles define aspects of clinical heterogeneity in attention deficit hyperactivity disorder. Nature Genetics . 2024;56:234–244. DOI: 10.1038/s41588-023-01593-7.
Cortese, S., Bellgrove, M. A., Brikell, I., Franke, B., Goodman, D. W., Hartman, C. A., et al. Attention-deficit/hyperactivity disorder (ADHD) in adults: evidence base, uncertainties and controversies. World Psychiatry . 2025;24(3):347–371. DOI: 10.1002/wps.21374.
Ogundele, M. O., et al. Subthreshold Autism and ADHD: A Brief Narrative Review for Frontline Clinicians. Pediatric Reports . 2025;17(2):42.
Fuermaier, A. B. M., Gontijo-Santos Lima, C., & Tucha, O. Impairment Assessment in Adult ADHD and Related Disorders: Current Opinions From Clinic and Research. Journal of Attention Disorders . 2024;28(12):1529–1541. DOI: 10.1177/10870547241261598.
Faraone, S. V., et al. Attention-deficit/hyperactivity disorder. Nature Reviews Disease Primers . 2024;10:11. DOI: 10.1038/s41572-024-00495-0.
Mattheisen, M., et al. Differences in the genetic architecture of common and rare variants in childhood, persistent and late-diagnosed attention-deficit hyperactivity disorder. Nature Genetics . 2022.
World Health Organization. ICD-11 for Mortality and Morbidity Statistics . Attention deficit hyperactivity disorder classification. Accessed 2026-08-17.
文獻使用聲明
本文僅使用上述研究建立截至 2026-08-17 的外部實證邊界。
本文提出的 CCSH、configuration space Ω C \Omega_C Ω C 、configuration density ρ ( c ) \rho(\mathbf c) ρ ( c ) 、clinical decision operator Γ \Gamma Γ 、subthreshold region R sub \mathcal R_{\text{sub}} R sub 、impairment surface Φ ( c , e ) \Phi(\mathbf c,\mathbf e) Φ ( c , e ) 、continuous-plus-cluster model 與 multi-layer geometry,均為本文理論構件,不應被誤認為上述研究作者的原始結論。
不同研究包含一般人口學童、成人 twins、clinical ADHD、pediatric neuroimaging cohorts、genetic cohorts、professional surveys 與 narrative reviews。本文不把它們視為單一大型實驗的直接累加證據。
狀態: v0.1,理論稿新增原始臨床/人體數據: 無醫學用途: 無下一篇: 《情境匹配與性能反轉:ADHD 配置何時成為障礙,何時可能成為優勢?》