Supplementary appendix

This appendix formed part of the original submission and has been peer reviewed.
We post it as supplied by the authors.

Supplement to: Li R, Li R, Xie J, et al. Associations of socioeconomic status and healthy

lifestyle with incident early-onset and late-onset dementia: a prospective cohort study.
Lancet Healthy Longev 2023; 4: e693–702.
 1 Associations of Socioeconomic Status and Healthy Lifestyle with Incident Early-

2 onset Dementia: A Prospective Cohort Study

3 Online Supplementary Material

4 Supplementary table 1. Ascertainment of dementia cases in the UK Biobank study.

5 Supplementary table 2. Baseline characteristics of participants in the study of

6 associations of socioeconomic status and healthy lifestyle score with incident late-onset

7 dementia according to socioeconomic status.

8 Supplementary table 3. Baseline characteristics of participants included or excluded

9 from analyses due to missing information.

10 Supplementary table 4. Associations of healthy lifestyle score with incident early-

11 onset dementia.

12 Supplementary table 5. Associations of each individual socioeconomic factor with

13 incident early-onset dementia and mediation proportion of socioeconomic inequity in

14 early-onset dementia risk attributed to lifestyle.

15 Supplementary table 6. Associations of socioeconomic status with incident early-

16 onset dementia: sensitivity analyses and mediation proportion of socioeconomic

17 inequity in early-onset dementia risk attributed to lifestyle.

18 Supplementary table 7. Associations of socioeconomic status with incident early-

19 onset dementia: subgroup analyses and mediation proportion of socioeconomic inequity

20 in early-onset dementia risk attributed to lifestyle.

1
21 Supplementary table 8. Associations of socioeconomic status with incident late-onset

22 dementia and mediation proportion of socioeconomic inequity in late-onset dementia

23 risk attributed to lifestyle.

24 Supplementary table 9. Associations of healthy lifestyle score with incident late-onset

25 dementia.

26 Supplementary table 10. Associations of healthy lifestyle score with incident late-

27 onset dementia by socioeconomic status.

28 Supplementary table 11. Joint associations of healthy lifestyle score and

29 socioeconomic status with incident late-onset dementia.

30 Supplementary table 12. Associations of socioeconomic status with incident late-

31 onset dementia and mediation proportion of socioeconomic inequity in late-onset

32 dementia risk attributed to lifestyle: sensitivity analyses.

33 Supplementary figure 1. Flowchart of the study.

34 Supplementary figure 2. Associations of healthy lifestyle score with incident early-

35 onset dementia by socioeconomic status.

36 Supplementary Methods

37 Assessment of socioeconomic status (SES)

38 Assessment of lifestyle factors

39 Ethnic background

2
40 Education qualification

41 Air pollution score (APS)

42 Blood tests and Genetic data

43 Model assumption

44 Multiple imputation

45 Ratio of hazard ratio (RHR)

3
46 Supplementary Table 1. Ascertainment of dementia cases in the UK Biobank study.

ICD-9 ICD-10 Self-reported Medications

Dementia 290.2, 290.3, 290.4, 291.2, A81.0, F00, F00.0, F00.1, F00.2, F00.9, 1263 20003: Prescriptions including
294.1, memantine, donepezil, galantamine
F01, F01.0, F01.1, F01.2, F01.3, F01.8, F01.9,
and rivastigmine
331.0, 331.1, 331.2, 331.5
F02, F02.0, F02.1, F02.2, F02.3, F02.4, F02.8,
F03, F05.1, F10.6,
G30, G30.0, G30.1, G30.8, G30.9,
G31.0, G31.1, G31.8, I67.3

47

4
48 Supplementary Table 2. Baseline characteristics of participants in the study of associations of socioeconomic status and healthy lifestyle score with incident
49 late-onset dementia according to socioeconomic status.

Late-onset dementia
Total population High SES Medium SES Low SES
P value
(n=294 133) (n=66 182) (n=72 143) (n=155 808)
Age, years 61.00 (57.00-65.00) 59.00 (56.00-63.00) 60.00 (56.00-64.00) 63.00 (59.00-66.00) <.001
Male 138 235 (47.00) 35 904 (54.25) 35 776 (49.59) 66 555 (42.72) <.001
White ethnicity or race 283 425 (96.36) 63 995 (96.70) 69 905 (96.90) 149 525 (95.97) <.001
Education levels <.001
High qualification 92 322 (31.39) 57 823 (87.37) 21 819 (30.24) 12 680 (8.14)
Intermediate qualifications 139 605 (47.46) 8 359 (12.63) 47 929 (66.44) 83 317 (53.47)
Low qualifications 62 206 (21.15) - 2 395 (3.32) 59 811 (38.39)
Occupation <.001
Employed 275 102 (93.53) 64 859 (98.00) 66 905 (92.74) 143 338 (92.00)
Unemployed 17 497 (5.95) 1 193 (1.80) 4 656 (6.45) 11 648 (7.48)
None of the above 1 534 (0.52) 130 (0.20) 582 (0.81) 822 (0.53)
Household income, pounds <.001
<18,000 75 836 (25.78) - - 75 836 (48.67)
18,000-51,999 147 466 (50.14) 25 771 (38.94) 58 067 (80.49) 63 628 (40.84)
>52,000 54 917 (18.67) 40 411 (61.06) 14 076 (19.51) 430 (0.28)
Do not know 15 914 (5.41) - - 15 914 (10.21)
Never smoking 150 709 (51.41) 37 797 (57.18) 37 863 (52.58) 75 049 (48.41) <.001
No excessive alcohol consumption 205 417 (70.49) 48 912 (74.24) 51 556 (71.94) 104 949 (68.21) <.001
Physical activity 226 840 (81.28) 52 665 (80.71) 56 764 (81.07) 117 411 (81.63) <.001
Healthy diet 160 450 (54.55) 37 404 (56.52) 39 387 (54.60) 83 659 (53.70) <.001
Social isolation 139 601 (47.46) 24 934 (37.67) 29 522 (40.92) 85 145 (54.65) <.001
Quality sleep 183 993 (62.55) 44 582 (67.36) 46 696 (64.73) 92 715 (59.51) <.001
Depressive symptom 13 355 (4.82) 1 718 (2.68) 2 491 (3.62) 9 146 (6.35) <.001
5
 APOE4 carriers 74 267 (25.81) 16 655 (25.68) 18 275 (25.87) 39 337 (25.85) 0.67
Air pollution <.001
High levels of air pollution 92 787 (31.55) 19 544 (29.53) 20 100 (27.86) 53 143 (34.11)
Medium levels of air pollution 87 260 (29.67) 17 197 (25.98) 21 636 (29.99) 48 427 (31.08)
Low levels of air pollution 114 086 (38.79) 29 441 (44.48) 30 407 (42.15) 54 238 (34.81)
BMI, kg/m2 <.001
<18.5 or >=30.0 74 476 (25.44) 12 400 (18.80) 17 301 (24.07) 44 775 (28.91)
18.5-24.9 89 589 (30.61) 24 415 (37.01) 22 479 (31.27) 42 695 (27.57)
25.0-29.9 128 653 (43.95) 29 147 (44.19) 32 105 (44.66) 67 401 (43.52)
HbA1c, mmol/mol 35.90 (33.50-38.50) 35.20 (33.00-37.60) 35.60 (33.30-38.10) 36.30 (33.90-39.10) <.001
HDL-C, mmol/L 1.41 (1.18-1.69) 1.45 (1.21-1.74) 1.42 (1.19-1.70) 1.39 (1.16-1.67) <.001
LDL-C, mmol/L 3.57 (2.97-4.18) 3.60 (3.04-4.17) 3.61 (3.03-4.20) 3.54 (2.91-4.17) <.001
Triglycerides, mmol/L 1.55 (1.11-2.20) 1.44 (1.03-2.06) 1.53 (1.09-2.19) 1.60 (1.15-2.26) <.001
Family history of dementia 40 888 (13.90) 9 996 (15.10) 10 253 (14.21) 20 639 (13.25) <.001
Medical history
Hearing impairment 131 628 (44.75) 28 056 (42.39) 31 671 (43.90) 71 901 (46.15) <.001
Hypertension 192 445 (65.43) 37 447 (56.58) 45 334 (62.84) 109 664 (70.38) <.001
Diabetes 20 981 (7.13) 2 968 (4.48) 4 319 (5.99) 13 694 (8.79) <.001
Cardiovascular diseases 30 876 (10.50) 4 302 (6.50) 5 977 (8.28) 20 597 (13.22) <.001

50 SES, socioeconomic status; APOE4, apolipoprotein E4; BMI, body mass index; HbA1c, Hemoglobin A1C; HDL-C, high-density lipoprotein cholesterol; LDL-C, low-
51 density lipoprotein cholesterol. Late-onset dementia was defined as dementia diagnosed ≥65 years, and the analyses were conducted among individuals aged 65

52 years and older at the end of follow-up.

53 266 443 individuals (92.40%) reached the endpoint, 21 661individuals (7.51%) were reported as deceased, and 261 participants (0.09%) were lost to follow-up.
54 P values were calculated using analysis of variance and χ2 test for continuous and categorical variables, respectively.

6
55 Supplementary table 3. Baseline characteristics of participants included or excluded from analyses due to missing information on socioeconomic factors.

Early-onset dementia Late-onset dementia

Characteristics Included Excluded Included Excluded
P value P value
(n=257 345) (n=27 061) (n=294 133) (n=44 388)
Age, years 51.00 (46.00-56) 52.00 (47.00-56.00) <.001 61.00 (57.00-65.00) 62.00 (59.00-66.00) <.001
Male 115 570 (44.91) 10 448 (38.61) <.001 138 235 (47.00) 16 877 (38.02) <.001
White ethnicity or race 239 467 (93.05) 22 768 (84.14) <.001 283 425 (96.36) 41 066 (92.52) <.001
Education levels <.001 <.001
High qualification 99 660 (38.73) 5 432 (25.13) 92 322 (31.39) 6 534 (17.55)
Intermediate qualifications 134 104 (52.11) 12 609 (58.34) 139 605 (47.46) 18 573 (49.89)
Low qualifications 23 581 (9.16) 3 571 (16.52) 62 206 (21.15) 12 121 (32.56)
Occupation <.001 <.001
Employed 228 002 (88.60) 20 718 (82.39) 275 102 (93.53) 39 299 (92.11)
Unemployed 27 944 (10.86) 4 081 (16.23) 17 497 (5.95) 2 991 (7.01)
None of the above 1 399 (0.54) 346 (1.38) 1 534 (0.52) 376 (0.88)
Household income, pounds <.001 <.001
<18,000 36 798 (14.30) 544 (37.23) 75 836 (25.78) 785 (45.32)
18,000-51,999 124 610 (48.42) 503 (34.43) 147 466 (50.14) 562 (32.45)
>52,000 87 233 (33.90) 92 (6.30) 54 917 (18.67) 79 (4.56)
Do not know 8 704 (3.38) 322 (22.04) 15 914 (5.41) 306 (17.67)
Never smoking 148 732 (57.93) 15 784 (61.68) <.001 150 709 (51.41) 23 515 (55.14) <.001
No excessive alcohol consumption 179 528 (70.40) 17 946 (69.49) 0.002 205 417 (70.49) 30 538 (71.38) <.001
Physical activity 197 835 (80.00) 16 384 (70.29) <.001 226 840 (81.28) 28 827 (74.90) <.001
Healthy diet 131 566 (51.13) 14 006 (52.94) <.001 160 450 (54.55) 24 699 (56.39) <.001
Social isolation 123 819 (48.11) 12 398 (45.96) <.001 139 601 (47.46) 19 587 (44.21) <.001
Quality sleep 162 842 (75.62) 13 469 (74.39) <.001 183 993 (62.55) 23 821 (53.67) <.001
Depressive symptom 17 463 (7.24) 2 018 (8.91) <.001 13 355 (4.82) 2 145 (5.51) <.001
APOE4 carriers 65 742 (26.15) 6 018 (25.04) <.001 74 267 (25.81) 10 343 (25.39) 0.07

7
 Air pollution <.001 <.001
High levels of air pollution 79 787 (31.00) 9 674 (35.75) 92 787 (31.55) 14 876 (33.51)
Medium levels of air pollution 75 639 (29.39) 8 180 (30.23) 87 260 (29.67) 13 579 (30.59)
Low levels of air pollution 101 919 (39.60) 9 207 (34.02) 114 086 (38.79) 15 933 (35.89)
BMI, kg/m2 <.001 <.001
<18.5 or >=30.0 62 626 (24.46) 6 986 (26.30) 74 476 (25.44) 11 669 (26.61)
18.5-24.9 89 696 (35.03) 9 037 (34.02) 89 589 (30.61) 13 261 (30.25)
25.0-29.9 103 747 (40.52) 10 541 (39.68) 128 653 (43.95) 18 915 (43.14)
HbA1c, mmol/mol 34.40 (32.10, 37.00) 34.90 (32.40, 37.50) <.001 35.90 (33.50, 38.50) 36.20 (33.80, 38.90) <.001
HDL-C, mmol/L 1.39 (1.17, 1.66) 1.40 (1.17, 1.68) 0.01 1.41 (1.18, 1.69) 1.43 (1.19, 1.71) <.001
LDL-C, mmol/L 3.51 (2.97, 4.08) 3.52 (2.97, 4.10) 0.26 3.57 (2.97, 4.18) 3.58 (2.96, 4.20) 0.68
Triglycerides, mmol/L 1.41 (0.98, 2.10) 1.42 (0.98, 2.10) 0.31 1.55 (1.11, 2.20) 1.54 (1.11, 2.18) 0.24
Family history of dementia 23 346 (9.07) 2 187 (8.08) <.001 40 888 (13.90) 5 501 (12.39) <.001
Medical history
Hearing impairment 93 936 (36.50) 9 258 (34.21) <.001 131 628 (44.75) 18 348 (41.34) <.001
Hypertension 116 434 (45.24) 12 897 (47.66) <.001 192 445 (65.43) 29 783 (67.10) <.001
Diabetes 11 440 (4.45) 1 471 (5.44) <.001 20 981 (7.13) 3 408 (7.68) <.001
Cardiovascular diseases 12 292 (4.78) 1 521 (5.62) <.001 30 876 (10.50) 5 031 (11.33) <.001

56 SES, socioeconomic status; APOE4, apolipoprotein E4; BMI, body mass index; HbA1c, Hemoglobin A1C; HDL-C, high-density lipoprotein cholesterol; LDL-C, low-
57 density lipoprotein cholesterol. Early-onset dementia was defined as dementia diagnosed <65 years, and the analyses were conducted among participants aged

58 younger than 60 years at baseline. Late-onset dementia was defined as dementia diagnosed ≥65 years, and the analyses were conducted among individuals aged
59 65 years and older at the end of follow-up.
60 The number of participants with missing values in the analyses of socioeconomic status and healthy lifestyle score with incident early-onset dementia were as
61 follows: education levels (5 449), occupation (1 916), and household income (25 600). The number of participants with missing values in the analyses of
62 socioeconomic status and healthy lifestyle score with incident late-onset dementia were as follows: education levels (7 160), occupation (1 722), and household

8
63 income (42 656).
64 P values were calculated using analysis of variance and χ2 test for continuous and categorical variables, respectively.

65

9
66 Supplementary table 4. Associations of healthy lifestyle score with incident early-onset dementia.

Hazard ratio (95% CI)

No. of events/person-
HLS No. of participants Unadjusted for socioeconomic Adjusted for socioeconomic
years
status status
Unfavorable 36 278 126/390 257 1 (Reference) 1 (Reference)
Intermediate 76 029 159/828 444 0.68 (0.54, 0.86) 0.71 (0.56, 0.90)
Favorable 145 038 217/1 599 484 0.55 (0.44, 0.70) 0.61 (0.49, 0.77)

67 HLS, healthy lifestyle score; CI, confidence interval.

68 Early-onset dementia was defined as dementia diagnosed <65 years, and the analyses were conducted among participants aged younger than 60
69 years at baseline. All models were included age (continuous), sex (male, female), self-reported race (white, non-white), study center (England,
70 Wales, Scotland), social isolation (yes, no), sleep quality (good quality, poor quality), depressive symptoms (yes, no), APOE4 carriers (yes, no),
71 body mass index (continuous), high-density lipoprotein cholesterol (continuous), low-density lipoprotein cholesterol (continuous), triglycerides
72 (continuous), HbA1c (continuous), prevalent comorbidities (hearing impairment, hypertension, and cardiovascular disease) (yes, no), and family
73 history of dementia (yes, no).

74

10
75 Supplementary table 5. Associations of each individual socioeconomic factor with incident early-onset dementia and mediation proportion of
76 socioeconomic inequity in early-onset dementia risk attributed to lifestyle. *

Hazard ratio (95% CI)

Mediation proportion
Factors Unadjusted for healthy Adjusted for healthy
(%) (95% CI)
lifestyle score lifestyle score
Education levels (low vs high qualification) 1.51 (1.14, 2.01) 1.46 (1.10, 1.94) 9.4 (2.9, 26.1)
Occupation (unemployment vs employment) 2.26 (1.81, 2.82) 2.19 (1.75, 2.75) 2.8 (1.5, 5.3)
Household income (<18 000 vs >52 000 pounds) 2.92 (2.15, 3.96) 2.89 (2.13, 3.92) 1.0 (0.4, 2.6)

77 CI, confidence interval.

78 * Early-onset dementia was defined as dementia diagnosed <65 years, and the analyses were conducted among participants aged younger than 60
79 years at baseline. All models were included all individual socioeconomic factors, as well as, age (continuous), sex (male, female), self-reported
80 race (white, non-white), study center (England, Wales, Scotland), social isolation (yes, no), sleep quality (good quality, poor quality), depressive
81 symptoms (yes, no), APOE4 carriers (yes, no), body mass index (continuous), high-density lipoprotein cholesterol (continuous), low-density
82 lipoprotein cholesterol (continuous), triglycerides (continuous), HbA1c (continuous), prevalent comorbidities (hearing impairment, hypertension,
83 and cardiovascular disease) (yes, no), and family history of dementia (yes, no).
84

11
85 Supplementary table 6. Associations of socioeconomic status with incident early-onset dementia and mediation proportion of
86 socioeconomic inequity in early-onset dementia risk attributed to lifestyle: sensitivity analyses. *

Hazard ratio (95% CI) Mediation

Analyses Unadjusted for healthy Adjusted for healthy proportion (%)
lifestyle score lifestyle score (95% CI)
Simultaneous adjustment for
3.50 (2.70, 4.53) 3.36 (2.59, 4.35) 4.4 (2.4, 8.0)
individual lifestyle factors
Weighted healthy lifestyle score 3.50 (2.70, 4.53) 3.38 (2.61, 4.37) 3.2 (1.8, 5.7)
Multiple imputation for education,
3.58 (2.72, 4.70) 3.41 (2.59, 4.50) 5.1 (3.0, 8.4)
income, and employment
Excluding participants with prevalent
3.39 (2.62, 4.39) 3.27 (2.52, 4.24) 3.3 (1.9, 5.9)
Parkinson’s disease
Excluding participants with prevalent
3.30 (2.54, 4.29) 3.19 (2.45, 4.15) 3.0 (1.6, 5.7)
Parkinson’s disease and stroke
Among participants aged 65 years old
3.22 (2.52, 4.11) 3.04 (2.37, 3.89) 3.5 (1.9, 6.3)
or younger
Excluding events that occurred within
3.56 (2.72, 4.66) 3.44 (2.62, 4.50) 3.3 (1.8, 5.9)
the first three years of follow-up
Included the Townsend deprivation
3.07 (2.36, 4.01) 3.00 (2.30, 3.91) 2.6 (1.3, 5.0)
index in model
Using socioeconomic score to
investigate more extreme 8.66 (6.03, 12.44) 8.14 (5.65, 11.72) 4.2 (1.9, 9.0)
socioeconomic disparity

87 CI, confidence interval.

12
88 * Early-onset dementia was defined as dementia diagnosed <65 years, and the analyses were conducted among participants aged younger
89 than 60 years at baseline. Only the results comparing the low with high socioeconomic status are reported. All models were included age
90 (continuous), sex (male, female), self-reported race (white, non-white), study center (England, Wales, Scotland), social isolation (yes, no),
91 sleep quality (good quality, poor quality), depressive symptoms (yes, no), APOE4 carriers (yes, no), body mass index (continuous), high-
92 density lipoprotein cholesterol (continuous), low-density lipoprotein cholesterol (continuous), triglycerides (continuous), HbA1c
93 (continuous), prevalent comorbidities (hearing impairment, hypertension, and cardiovascular disease) (yes, no), and family history of
94 dementia (yes, no).

13
95 Supplementary table 7. Associations of socioeconomic status with incident early-
96 onset dementia and mediation proportion of socioeconomic inequity in early-onset
97 dementia risk attributed to lifestyle: subgroup analyses. *

Hazard ratio (95% CI)

Mediation
Unadjusted for Adjusted for
Analyses proportion (%)
healthy lifestyle healthy lifestyle
(95% CI)
score score
Sex
Female 2.89 (1.98, 4.21) 2.85 (1.95, 4.16) <1
Male 4.05 (2.84, 5.78) 3.81 (2.67, 5.45) 6.0 (3.4, 10.4)
Race
White 3.43 (2.64, 4.47) 3.30 (2.54, 4.29) 3.6 (2.1, 6.3)
Non-white 11.29 (1.52, 83.68) 11.33 (1.53, 84.04) <1
Apolipoprotein E4
carriers
Yes 3.86 (2.50, 5.96) 3.77 (2.44, 5.82) 2.1 (0.7, 5.8)
No 3.30 (2.39, 4.55) 3.16 (2.29, 4.36) 4.0 (2.0, 7.9)
Family history of
dementia
Yes 2.21 (1.16, 4.20) 2.10 (1.10, 3.99) 4.7 (0.8, 24.2)
No 3.80 (2.86, 5.05) 3.68 (2.77, 4.89) 3.1 (1.7, 5.6)
Hearing problem
Yes 3.29 (2.28, 4.75) 3.22 (2.22, 4.66) 2.6 (1.0, 6.9)
No 3.77 (2.63, 5.42) 3.58 (2.49, 5.14) 3.9 (2.0, 7.6)
Hypertension
Yes 4.13 (2.83, 6.03) 4.03 (2.76, 5.89) 1.9 (0.7, 5.3)
No 2.95 (2.06, 4.24) 2.80 (1.95, 4.03) 5.4 (2.7, 10.6)
Diabetes
Yes 11.05 (2.66, 45.86) 11.21 (2.7, 46.57) <1
No 3.27 (2.50, 4.26) 3.12 (2.39, 4.08) 4.1 (2.4, 6.8)
Air pollution score
Highest tertile 3.22 (2.14, 4.89) 3.16 (2.10, 4.76) 2.2 (0.6, 7.8)
Lowest tertile 3.00 (1.90, 4.71) 2.88 (1.83, 4.53) 3.0 (1.0, 8.3)
98 * Early-onset dementia was defined as dementia diagnosed <65 years, and the analyses
99 were conducted among participants aged younger than 60 years at baseline. Only the
100 results comparing the low with high socioeconomic status are reported. All models were
101 included age (continuous), sex (male, female), self-reported race (white, non-white),
102 study center (England, Wales, Scotland), social isolation (yes, no), sleep quality (good
103 quality, poor quality), depressive symptoms (yes, no), APOE4 carriers (yes, no), body
14
104 mass index (continuous), high-density lipoprotein cholesterol (continuous), low-
105 density lipoprotein cholesterol (continuous), triglycerides (continuous), HbA1c
106 (continuous), prevalent comorbidities (hearing impairment, hypertension, and
107 cardiovascular disease) (yes, no), and family history of dementia (yes, no).

15
108 Supplementary table 8. Associations of socioeconomic status with incident late-onset dementia and mediation proportion of socioeconomic
109 inequity in late-onset dementia risk attributed to lifestyle. *
Hazard ratio (95% CI) Mediation
No. of No. of
SES Unadjusted for healthy Adjusted for healthy proportion (%)
participants events/person-years
lifestyle score lifestyle score (95% CI)
High SES 66 182 592/831 266 1 (Reference) 1 (Reference)
Medium SES 72 143 894/901 129 1.17 (1.05, 1.29) 1.16 (1.05, 1.29) 1.5 (0.3, 6.1)
Low SES 155 808 4 282/1 904 823 1.71 (1.57, 1.87) 1.70 (1.55, 1.86) 1.9 (1.2, 3.0)

110 SES, socioeconomic status; CI, confidence interval.

111 *Late-onset dementia was defined as dementia diagnosed ≥65 years, and the analyses were conducted among individuals aged 65 years and older
112 at the end of follow-up. All models were included age (continuous), sex (male, female), self-reported race (white, non-white), study center (England,
113 Wales, Scotland), social isolation (yes, no), sleep quality (good quality, poor quality), depressive symptoms (yes, no), APOE4 carriers (yes, no),
114 body mass index (continuous), high-density lipoprotein cholesterol (continuous), low-density lipoprotein cholesterol (continuous), triglycerides
115 (continuous), HbA1c (continuous), prevalent comorbidities (hearing impairment, hypertension, and cardiovascular disease) (yes, no), and family
116 history of dementia (yes, no).

16
117 Supplementary table 9. Associations of healthy lifestyle score with incident late-onset dementia. *

Hazard ratio (95% CI)

No. of No. of events/person-
HLS Unadjusted for Adjusted for
participants years
socioeconomic status socioeconomic status
Unfavorable 42 044 905/509 810 1 (Reference) 1 (Reference)
Intermediate 89 272 1 874/1 097 399 0.93 (0.85, 1.00) 0.93 (0.86, 1.01)
Favorable 162 817 2 989/2 030 009 0.83 (0.77, 0.90) 0.85 (0.79, 0.92)

118 HLS, healthy lifestyle score; CI, confidence interval.

119 *Late-onset dementia was defined as dementia diagnosed ≥65 years, and the analyses were conducted among individuals aged 65 years and older
120 at the end of follow-up. All models were included age (continuous), sex (male, female), self-reported race (white, non-white), study center (England,
121 Wales, Scotland), social isolation (yes, no), sleep quality (good quality, poor quality), depressive symptoms (yes, no), APOE4 carriers (yes, no),
122 body mass index (continuous), high-density lipoprotein cholesterol (continuous), low-density lipoprotein cholesterol (continuous), triglycerides
123 (continuous), HbA1c (continuous), prevalent comorbidities (hearing impairment, hypertension, and cardiovascular disease) (yes, no), and family
124 history of dementia (yes, no).

125

126

17
127 Supplementary table 10. Associations of healthy lifestyle score with incident late-onset dementia by socioeconomic status. *

No. of events/person-
Subgroup No. of participants Hazard ratio (95% CI) P value
years
High SES
0 or 1 healthy behavior 7 889 68/98 317 1 (Reference)
2 healthy behaviors 18 429 174/230 669 1.07 (0.81, 1.42) 0.64
3 or 4 healthy behaviors 39 864 350/502 279 1.02 (0.79, 1.33) 0.87
Medium SES
0 or 1 healthy behavior 9 898 124/122 516 1 (Reference)
2 healthy behaviors 21 622 282/269 075 0.95 (0.77, 1.18) 0.67
3 or 4 healthy behaviors 40 623 488/509 539 0.88 (0.72, 1.08) 0.23
Low SES
0 or 1 healthy behavior 24 257 713/288 977 1 (Reference)
2 healthy behaviors 49 221 1 418/597 655 0.92 (0.84, 1.00) 0.06
3 or 4 healthy behaviors 82 330 2 151/1 018 191 0.83 (0.76, 0.90) <.001

128 SES, socioeconomic status; CI, confidence interval. *Late-onset dementia was defined as dementia diagnosed ≥65 years, and the analyses were
129 conducted among individuals aged 65 years and older at the end of follow-up. All models were included age (continuous), sex (male, female), self-
130 reported race (white, non-white), study center (England, Wales, Scotland), social isolation (yes, no), sleep quality (good quality, poor quality),
131 depressive symptoms (yes, no), APOE4 carriers (yes, no), body mass index (continuous), high-density lipoprotein cholesterol (continuous), low-
132 density lipoprotein cholesterol (continuous), triglycerides (continuous), HbA1c (continuous), prevalent comorbidities (hearing impairment,
133 hypertension, and cardiovascular disease) (yes, no), and family history of dementia (yes, no).

18
134 Supplementary table 11. Joint associations of healthy lifestyle score and socioeconomic status with incident late-onset dementia. *

Subgroup No. of participants No. of events/person-years Hazard ratio (95% CI) P value
High SES
3 or 4 healthy behaviors 39 864 350/502 279 1 (Reference)
2 healthy behaviors 18 429 174/230 669 1.05 (0.87, 1.26) 0.62
0 or 1 healthy behavior 7 889 68/98 317 0.98 (0.75, 1.27) 0.86
Medium SES
3 or 4 healthy behaviors 40 623 488/509 539 1.14 (0.99, 1.31) 0.07
2 healthy behaviors 21 622 282/269 075 1.23 (1.05, 1.44) 0.01
0 or 1 healthy behavior 9 898 124/122 516 1.29 (1.05, 1.58) 0.02
Low SES
3 or 4 healthy behaviors 82 330 2 151/1 018 191 1.62 (1.45, 1.82) <.001
2 healthy behaviors 49 221 1 418/597 655 1.80 (1.60, 2.03) <.001
0 or 1 healthy behavior 24 257 713/288 977 1.97 (1.73, 2.24) <.001

135 SES, socioeconomic status; CI, confidence interval. *Late-onset dementia was defined as dementia diagnosed ≥65 years, and the analyses were
136 conducted among individuals aged 65 years and older at the end of follow-up. All models were included age (continuous), sex (male, female), self-
137 reported race (white, non-white), study center (England, Wales, Scotland), social isolation (yes, no), sleep quality (good quality, poor quality),
138 depressive symptoms (yes, no), APOE4 carriers (yes, no), body mass index (continuous), high-density lipoprotein cholesterol (continuous), low-
139 density lipoprotein cholesterol (continuous), triglycerides (continuous), HbA1c (continuous), prevalent comorbidities (hearing impairment,
140 hypertension, and cardiovascular disease) (yes, no), and family history of dementia (yes, no).
141

19
142 Supplementary table 12. Associations of socioeconomic status with incident late-onset dementia and mediation proportion of
143 socioeconomic inequity in late-onset dementia risk attributed to lifestyle: sensitivity analyses. *

Hazard ratio (95% CI) Mediation

Analyses Unadjusted for healthy Adjusted for healthy proportion (%)
lifestyle score lifestyle score (95% CI)
Simultaneous adjustment for
1.71 (1.57, 1.87) 1.70 (1.55, 1.86) 1.5 (0.7, 3.3)
individual lifestyle factors
Weighted healthy lifestyle score 1.71 (1.57, 1.87) 1.71 (1.56, 1.86) 1.4 (1.0, 2.1)
Multiple imputation for education,
1.65 (1.51, 1.80) 1.64 (1.50, 1.79) 1.6 (1.0, 2.6)
income, and employment
Excluding participants with prevalent
1.76 (1.60, 1.92) 1.74 (1.59, 1.91) 1.4 (0.8, 2.3)
Parkinson’s disease
Excluding participants with prevalent
1.75 (1.59, 1.92) 1.73 (1.58, 1.90) 1.3 (0.8, 2.3)
Parkinson’s disease and stroke
Excluding events that occurred within
1.70 (1.55, 1.86) 1.69 (1.54, 1.84) 1.5 (0.9, 2.5)
the first three years of follow-up
Included the Townsend deprivation
1.66 (1.52, 1.81) 1.65 (1.51, 1.80) 1.1 (0.6, 2.0)
index in model
Using socioeconomic score to
investigate more extreme 2.74 (2.26, 3.33) 2.69 (2.22, 3.27) <1
socioeconomic disparity

144 CI, confidence interval.

145 * Late-onset dementia was defined as dementia diagnosed ≥65 years, and the analyses were conducted among individuals aged 65 years

20
146 and older at the end of follow-up. Only the results comparing the low with high socioeconomic status are reported. All models were
147 included age (continuous), sex (male, female), self-reported race (white, non-white), study center (England, Wales, Scotland), social
148 isolation (yes, no), sleep quality (good quality, poor quality), depressive symptoms (yes, no), APOE4 carriers (yes, no), body mass index
149 (continuous), high-density lipoprotein cholesterol (continuous), low-density lipoprotein cholesterol (continuous), triglycerides
150 (continuous), HbA1c (continuous), prevalent comorbidities (hearing impairment, hypertension, and cardiovascular disease) (yes, no), and
151 family history of dementia (yes, no).
152

21
153

154 Supplementary figure 1. Flowchart of the study.

22
155

156 Supplementary figure 2. Associations of healthy lifestyle score with incident early-onset dementia by socioeconomic status. *

157 SES, socioeconomic status; CI, confidence interval; RERI, relative excess risk due to interaction; AP, attributable proportion due to interaction.
158 * Early-onset dementia was defined as dementia diagnosed <65 years, and the analyses were conducted among participants aged younger than 60

23
159 years at baseline. Hazard ratios were adjusted for age (continuous), sex (male, female), self-reported race (white, non-white), study center (England,
160 Wales, Scotland), social isolation (yes, no), sleep quality (good quality, poor quality), depressive symptoms (yes, no), APOE4 carriers (yes, no),
161 body mass index (continuous), high-density lipoprotein cholesterol (continuous), low-density lipoprotein cholesterol (continuous), triglycerides
162 (continuous), HbA1c (continuous), prevalent comorbidities (hearing impairment, hypertension, and cardiovascular disease) (yes, no), and family
163 history of dementia (yes, no). Squares indicated hazard ratios and horizontal lines denoted the ranges of the 95% CIs. The vertical dash lines
164 indicated the hazard ratio of 1.00. Multiplicative interaction was evaluated using hazard ratios for the product term between the healthy lifestyle
165 score (0 or 1 point v 3 or 4 points) and SES (low v high). Additive interaction was evaluated using RERI and AP between the healthy lifestyle
166 score (0 or 1 point v 3 or 4 points) and SES (low v high).
167

24
168 Supplementary Methods

169 Assessment of socioeconomic status (SES)

170 Education qualification, average household income, and employment status were used

171 to construct an overall SES variable using latent class analysis. The UK Biobank

172 collected total household income before tax using questionnaires (<₤18 000, ₤18 000-

173 £30 999, ₤31 000-£51 999, ₤52 000-£100 000, >₤100 000, do not know, or prefer not to

174 answer). Education qualifications were reported by participants and were classified as

175 college or university degree; A levels, AS levels, or equivalent; O levels, GCSEs, or

176 equivalent; CSEs or equivalent; NVQ, HND, HNC, or equivalent; other professional

177 qualifications; none of the above; or prefer not to answer. Current employment status

178 was acquired as in paid employment or self-employed; retired; looking after home;

179 unable to work because of sickness or disability; doing unpaid or voluntary work; being

180 full or part time students; none of the above; or prefer not to answer. We exclude

181 participants choosing the last options (prefer not to answer) as missing values in the

182 main analyses. A total of 55 817 participants excluded (11.1% among total participants)

183 had missing information about total household income, while other (2 702 participants)

184 had missing information about education qualifications and employment status. The

185 method for constructing SES with latent class analysis has been described in detail in a

186 previous study1. In brief, the SAS process, PROC LCA, was used to calculate the

187 Akaike information criterion (AIC), Bayesian information criterion (BIC), likelihood

188 ratio statistic G2, and item-response probability. The mean posterior probability

25
189 reflected the uncertainty of posterior classification and a value of 0.7 or more indicated

190 an acceptable uncertainty.

191

26
192 Since the model with four latent classes failed to converge, three-latent class solution were conducted. In the three-latent class solution, the

193 G2 is 3 340, AIC is 3 422, BIC is 3 873, and mean posterior probabilities of all latent classes were above 0.70 (Supplementary table 13). As

194 shown below (Supplementary table 13, and Supplementary figure 3), latent class 1 was characterized by low-level average household income,

195 low education qualifications, and unemployment, which could be defined as “low SES”; latent class 2 was characterized by high-level average

196 household income, high education qualifications, and employment, which could be defined as “high SES”; latent class 3 was characterized by

197 medium- level average household income, high and medium education qualification, and employment, which could be defined as “medium SES”.

198 Supplementary table 13. Mean posterior probabilities, prevalence of latent classes, and item-related proportion of each latent class.

Item Latent class 1 Latent class 2 Latent class 3

Mean posterior probability 0.90 0.77 0.72
Prevalence 0.31 0.19 0.50
Education qualification
College or university degree 0.13 0.78 0.31
A levels, AS levels, or equivalent 0.07 0.17 0.12
O levels, GCSEs, or equivalent 0.19 0.02 0.30
CSEs or equivalent 0.05 <0.01 0.07
NVQ, HND, HNC, or equivalent 0.07 0.01 0.09
Other professional qualifications 0.05 0.01 0.07
None of the above 0.45 0.00 0.05

27
 Average household income
Less than ₤18 000 0.69 0.00 0.00
₤18 000-£30 999 0.15 0.00 0.39
₤31 000-£51 999 <0.01 0.01 0.49
₤52 000-£100 000 0.00 0.71 0.12
More than ₤100 000 <0.01 0.27 0.00
Do not know 0.15 0.00 0.00
Employment status
Employed 0.81 0.94 0.98
Unemployed 0.18 0.06 0.02
None of the above 0.01 <0.01 <0.01
199

200

201 Supplementary figure 3. Practical definitions of high, medium, and low socioeconomic status.

28
202 In sensitivity analysis, a socioeconomic score was constructed on a scale of 0 to 6. This

203 score was determined based on three key factors: education qualification, average household

204 income, and employment status. Each factor was assigned a point value of 0, 1, or 2,

205 corresponding to low, medium, or high levels.

206 For education qualification, a high level represented a college or university degree; a

207 medium level encompassed A levels, AS levels, or their equivalents, O levels, GCSEs, or their

208 equivalents, CSEs or equivalents, NVQ, HND, HNC, or their equivalents, as well as other

209 professional qualifications; and a low level represented cases that no qualifications of the above

210 were attained. Regarding average household income, a high level denoted an income of

211 ≥₤52,000; a medium level covered the range of ₤18,000 to £51,999; and a low level

212 encompassed incomes below ₤18,000, along with cases where the income level was unknown.

213 In terms of employment status, a high level indicated being employed; and a low level indicated

214 being unemployed.

215 Assessment of lifestyle factors

216 Information on lifestyle factors was self-reported by participants through questionnaires.

217 Healthy lifestyle score was conducted based on four lifestyle factors, including cigarette

218 smoking, alcohol consumption, physical activity, and diet (Supplementary table 14). Never

219 smoking was defined as smoking fewer than 100 cigarettes in life. Participants self-reported

220 frequency and volume of current alcohol consumption, and daily consumption of one drink or

221 fewer for women and two drinks or fewer for men was defined as a healthy level, according to

222 the dietary guidelines (one drink contains 8 g of ethanol)2. Weekly metabolic equivalent hours

29
 223 of time physical activity were calculated, we further defined a healthy level of physical activity

224 according to WHO recommendation. According to a previous UK Biobank study, we evaluated

225 dietary quality using a more recent dietary recommendation for cardiovascular health, which

226 considered adequate consumption of fruit, vegetables, whole grains, and fish and reduced

227 consumption of refined grains, processed meats, and unprocessed meats3.

228 For each lifestyle factor, we assigned 1 point for a healthy level and 0 points for an

229 unhealthy level. Thus, the healthy lifestyle score was the sum of the points and ranged between

230 0 and 4, with higher scores indicating healthier lifestyles. We divided the participants into three

231 groups according to healthy lifestyle score, i.e., unfavorable (0-1 point), intermediate (2 points),

232 and unfavorable (3-4 points).

233 In sensitivity analysis, we also constructed a weighted lifestyle score by summing up

234 weighted healthy lifestyle factors (cigarette smoking, alcohol consumption, physical activity,

235 and diet). Weighted coefficients (β coefficients) were extracted from the multivariable-adjusted

236 risk estimates of single healthy lifestyle factors. The equation was: weighted lifestyle score =

237 (βcigarette smoking × cigarette smoking + βalcohol consumption × alcohol consumption + βphysical activity ×

238 physical activity + βdiet ×diet) × (4 / sum of β coefficients).

239 Supplementary table 14. Ascertainment of healthy lifestyles.

Lifestyle factor UK Biobank field code Healthy level

Cigarette smoking 1239, 1249, 2644 Smoke fewer than 100 cigarettes in life.
Alcohol 1558, 1568, 1578, 1588, 1598, 1608, One drink or fewer for women and two drinks or
consumption 5364, 4407, 4418, 4429, 4440, 4451, fewer for men per day.
4462

30
Physical activity 884, 894, 904, 914 More than 150 minutes of moderate physical activity
per week, 75 minutes of vigorous physical activity
per week, or equivalent.
Fruit 1309, 1319 ≥3 servings/day
Vegetables 1289, 1299 ≥3 servings/day
Fish 1329, 1339 ≥2 servings/day
Whole grain 1438, 1448, 1458, ≥3 servings/day
Meeting at least 3 of the
Diet 1468
dietary recommendation.
Refined grain 1438, 1448, 1458, ≤1.5 servings/week
1468
Red meat 1369, 1379, 1389 ≤1.5 servings/week
Processed meat 1349 ≤1 serving/week

240 Ethnic background

241 In the UK biobank, ethnic background included White, Mixed (White and Black Caribbean,

242 White and Black African, White and Asian, and Any other mixed background), Asian or Asian

243 British, Black or Black British, Chinese, and other ethnic group. In this study, we recorded race

244 into two categories, since 93% of total participants identified to be White race/ethnicity.

245 Education qualification

246 The education sector in the UK offers a diverse array of qualification types. One of the key

247 frameworks that govern these qualifications is the Regulated Qualifications Framework (RQF),

248 which replaced the Qualifications and Credit Framework on 1 October 2015. The RQF serves

249 as the backbone for accrediting qualifications in England, Northern Ireland, and Wales. Within

250 the RQF, qualifications are categorized into different levels based on their complexity, content,

251 and depth of study. These levels range from entry-level to level 8. These levels encompass a
31
252 wide range of academic and vocational achievements. For clarity, higher education

253 qualifications align with levels 4 to 8 of the RQF.

254 Supplementary table 15. The Regulated Qualifications Framework (RQF) in the United

255 Kingdom.

RQF Example qualification

8 Doctoral degree
7 Master's degree
6 Bachelor's degree
5 Higher National Diploma (HND)
4 Higher National Certificate (HNC)
3 A-level, National Diploma
2 General Certificate of Secondary Education (GCSE, grades A–C)
1 General Certificate of Secondary Education (GCSE, grades D-G)
Entry Skills for Life

256 Air pollution score (APS)

257 European Study of Cohorts for Air Pollution Effects (ESCAPE) project developed land-use

258 regression (LUR) models to estimate ambient air pollutants in 2010, including particle matter

259 with aerodynamic diameters of <2.5 μm (PM2.5), 2.5-10 μm (PMcoarse), and <10 μm (PM10)4;

260 nitrogen oxide (NOx) and nitrogen dioxide (NO2)5. The spatial variation of annual average air

261 pollutant concentrations was calculated by the LUR models including multiple geospatial

262 predictor variables generated from the Geographic Information System (such as topography,

263 traffic intensity, population, and land use)5. Favorable model performance for LUR-derivated

264 air pollutant concentrations (cross-validation R2 >75%) was shown in the leave-one-out cross-

265 validation5. Individual air pollutant exposures of participants from the UK Biobank were linked

266 to participants’ residential addresses provided at baseline. The representative air pollution

267 estimated in 2010 was analyzed in this study6.

32
268 APS was calculated by summing up weighted concentrations of air pollution (PM2.5,

269 PMcoarse, PM10, NO2, and NOx)7. Weighted coefficients (β coefficients) were extracted from the

270 multivariable-adjusted risk estimates of single air pollutants. The equation was: APS = (βPM2.5

271 × PM2.5 + βPMcoarse × PMcoarse + βPM10 × PM10 + βNO2 × NO2 +βNOx × NOx) × (5 / sum of β

272 coefficients). The APS ranged from 41.22 to 183.99, with a higher score representing a higher

273 level of combined exposure to air pollution. Additionally, we divided the APS into three levels

274 as high APS (the highest tertile), medium APS, and low APS (the lowest tertile) for categorical

275 analysis.

276 Blood tests and Genetic data

277 Blood samples were collected, analyzed, and quality controlled by the UK Biobank team

278 (https://biobank.ctsu.ox.ac.uk/crystal/ukb/docs/serum_biochemistry.pdf). The UK Biobank

279 team quality controlled and imputed genetic data which was genotyped by Affymetrix8. We use

280 two single-nucleotide polymorphisms (rs7412 and rs429358) to define the presence of APOE4)

281 with E3/E4 or E4/E4 defined as the presence of APOE49.

282 Model assumption

283 We examined the proportional hazards assumption by creating a product term of follow-up

284 time and exposure variables, and no significant deviation from the assumption was found. In

285 addition, the proportional hazards assumption was assessed by plotting the log-log survival

286 curve, the potential outliers were examined by visualizing the deviance residuals, and the

287 assumption that continuous covariates had a linear form was tested by plotting Martingale

288 residuals against continuous covariates. No assumptions were violated in the graphical

33
289 diagnostics.

290 Multiple imputation

291 We imputed missing data using multiple imputation (PROC MI) by chained equations with 5

292 imputations. Multiple imputation utilized appropriate statistical models to generate plausible

293 distributions of values to replace missing data. Continuous variables were imputed with linear

294 regression and categorical variables with logistic regression.

295 Ratio of hazard ratio (RHR)

296 To quantify the difference between the hazard ratios of early-onset dementia (EOD) and late-

297 onset dementia (LOD) with risk factors, we calculated a ratio of the hazard ratio as has been

298 used elsewhere10, 11. The equations were as follows:

299 RHR =𝑒 ( ln(HREOD ) − ln(HRLOD ) )

2 2
ln(𝑈𝐸𝑂𝐷 )−ln(𝐿𝐸𝑂𝐷 ) ln(𝑈𝐿𝑂𝐷 )−ln(𝐿𝐿𝑂𝐷 )
(ln(HREOD ) − ln(HRLOD )) ± (1.96 × √( ) +( ) )
2×1.96 2×1.96
300 RHR 95% CI =𝑒

301 HREOD, the HRs for EOD with risk factors; HRLOD, the HRs for LOD with risk factors; UEOD,

302 the upper confidence limit of the HRs for EOD with risk factors; LEOD, the lower confidence

303 limit of the HRs for EOD with risk factors; ULOD, the upper confidence limit of the HRs for

304 LOD with risk factors; LLOD, the lower confidence limit of the HRs for LOD with risk factors.

305

34
306 Reference:
307 1. Zhang YB, Chen C, Pan XF, et al. Associations of healthy lifestyle and socioeconomic status with
308 mortality and incident cardiovascular disease: two prospective cohort studies. BMJ (Clinical research
309 ed). 2021;373:n604.
310 2. NH S. The risks of drinking too much. 2022.
311 3. Lourida I, Hannon E, Littlejohns TJ, et al. Association of Lifestyle and Genetic Risk With
312 Incidence of Dementia. Jama. 2019;322:430-437.
313 4. Eeftens M, Beelen R, de Hoogh K, et al. Development of Land Use Regression models for PM(2.5),
314 PM(2.5) absorbance, PM(10) and PM(coarse) in 20 European study areas; results of the ESCAPE
315 project. Environmental science & technology. 2012;46:11195-205.
316 5. Beelen R, Hoek G, Vienneau D, et al. Development of NO2 and NOx land use regression models
317 for estimating air pollution exposure in 36 study areas in Europe–The ESCAPE project. Atmospheric
318 Environment. 2013;72:10-23.
319 6. Li X, Wang M, Song Y, et al. Obesity and the relation between joint exposure to ambient air
320 pollutants and incident type 2 diabetes: A cohort study in UK Biobank. PLoS medicine.
321 2021;18:e1003767.
322 7. Wang M, Zhou T, Song Y, et al. Joint exposure to various ambient air pollutants and incident heart
323 failure: a prospective analysis in UK Biobank. European heart journal. 2021;42:1582-1591.
324 8. Bycroft C, Freeman C, Petkova D, et al. The UK Biobank resource with deep phenotyping and
325 genomic data. Nature. 2018;562:203-209.
326 9. Saddiki H, Fayosse A, Cognat E, et al. Age and the association between apolipoprotein E genotype
327 and Alzheimer disease: A cerebrospinal fluid biomarker-based case-control study. PLoS medicine.
328 2020;17:e1003289.
329 10. Huxley RR and Woodward M. Cigarette smoking as a risk factor for coronary heart disease in
330 women compared with men: a systematic review and meta-analysis of prospective cohort studies.
331 Lancet. 2011;378:1297-305.
332 11. Batty GD, Gale CR, Kivimäki M, et al. Comparison of risk factor associations in UK Biobank
333 against representative, general population based studies with conventional response rates: prospective
334 cohort study and individual participant meta-analysis. BMJ (Clinical research ed). 2020;368:m131.
335

35