Stata 计算 患病率比（prevalence ratio）

*** M-H relative risk
5 T0 c- @& }3 P) D$ A

1. cs ill exposed [fweight = freq], by(confounder)

复制代码

*** Poisson regression unadjusted

1. poisson ill exposed confounder [fweight = freq], irr

复制代码

*** Poisson regression adjusted by chi-squared

1. glm ill exposed confounder [fweight = freq], family(poisson) scale(x2) eform

复制代码

*** Poisson regression adjusted by deviance

1. glm ill exposed confounder [fweight = freq], family(poisson) scale(dev) eform irls

复制代码

; f+ F' g2 [# S, z0 k8 R6 H& O*** Poisson regression with robust variance
! j2 J/ r, u- g8 g  V

1. poisson ill exposed confounder [fweight = freq], irr r

复制代码

/ H' F" f4 J2 C( Z6 J2 c& ~*** Log-binomial regression

1. glm ill exposed confounder [fweight = freq], family(binomial) link(log) eform

复制代码

*** Odds ratio from logistic regression
9 h8 N: d8 c" |% K, j, ?

1. logistic ill exposed confounder [fweight = freq]

复制代码

& X) I) O6 s% u3 S4 Q
* f6 B$ R9 }* F. tA comparison of two methods for estimating prevalence ratios.
; J1 q+ ^. S9 [. T+ J- Khttps://www.ncbi.nlm.nih.gov/m/p ... m=/14567763/related
: ~2 j9 N* E0 u8 K3 T' {! U" s
( G: h# X- X0 c) u9 h, M4 nPoisson regression with or without robust
* Q' C- W- G8 n# v. Q" E. xhttps://www.statalist.org/forums ... ussion/general/5912
9 }% u! H9 Q) k" I" i: a
Poisson Regression with Robust Variance in National Survey Data
https://www.statalist.org/forums ... ational-survey-data

Alternatives for logistic regression in cross-sectional studies: an empirical comparison of models that directly estimate the prevalence ratio
https://www.statalist.org/forums/filedata/fetch?id=5927
' g; X$ m1 z. u! ]) R- O, i* Khttps://www.ncbi.nlm.nih.gov/pubmed/14567763

8 U+ {7 o4 s# ]' u( g$ c

Task 1c: How to Generate Age-Adjusted Proportions or Prevalence Rates and Means Using Stata
https://www.cdc.gov/nchs/tutoria ... dization/Task1c.htm
6 n5 G, c" i7 n) g' P* q( Z
, _% @3 @. x6 B9 d! |+ f, `" @* W1 @
* ~& [! e& t: W: ZEstimating predicted probabilities from logistic regression: different methods correspond to different target populations
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4052139/

Marginal standardization is the appropriate method when making inference to the overall population.Other methods should be used with caution, and prediction at the means should not be used with binary confounders. Stata, but not SAS, incorporates simple methods for marginal standardization.
5 K. t8 x" D  y4 X0 b6 L5 n5 V4 [$ S
  w4 g: \, {- H% V4 d: u

1. * Logistic regression adjusting for confounders
   
2. * Outcome is “bmi3” – overweight/obese BMI at wave 3
   
3. * Exposure is low physical activity (inactive = 1)
   
4. 
   8 H/ h; Z/ H8 z; Y+ j
5. ** NOTE: Commands given separately for estimating prevalence ratio in v11 **
   
6. 
   
7. ****************************************************************************
   , q) A3 R; i) y7 }/ e
8. *** Method 1 – Marginal standardization (standard = TOTAL population)
   & R9 j! }6 y/ j  }9 J$ z
9. 
   ! T3 }$ }3 a5 O" C* p! V
10. * First run logistic regression
    
11. logit bmi3 i.inactive age18 female white diet bmi2, or
    + f' g4 Z0 ~" K+ ^7 h; ^' I
12. 
    
13. ** Estimate prevalence difference for inactive = 1 vs. inactive = 0
    * D  k1 k! S/ }$ j, S! V. ]
14. margins, dydx(inactive)                               
    * A( `6 d' ^$ a+ G0 E' G+ Y  f1 }# b
15. 
    - a+ g" [5 c' |" }* _
16. ** Estimate predicted probabilities by exposure level
    
17. margins inactive, post
    
18. 
    
19. ** Estimate prevalence ratio for inactive = 1 vs. inactive = 0
    $ u& |$ [- w+ ^& r+ ^: q9 u
20. ***************************
    ( E5 p+ j# {8 h5 z
21. *** FOR STATA 12 AND 13 ***
    9 U# l! ]( q& K! x' g) W8 i
22. margins, coeflegend
    
23. nlcom (ratio1: _b[1.inactive]/_b[0bn.inactive]), post       
    8 i( b; z( D$ m9 S' k; ?
24. test ratio1 = 1                                                       
    
25.         *** Gives appropriate p value for comparing ratio1 to 1
    
26. 
    * g- e; q9 R; z* m1 V2 I% F
27. * Alternative syntax
    6 W' i' [: Q, J) b0 Q
28. logit bmi3 i.inactive age18 female white diet bmi2, or
    + O  V$ R) w2 |
29. margins inactive, post
    / T3 ^+ O# L4 X  p/ b/ J* X; Y
30. margins, coeflegend
    5 `/ ~& E# P$ D; c) _' T
31. nlcom (ratio_ln: ln(_b[1.inactive]) - ln(_b[0bn.inactive] )), post
    3 T' A* W) Y) }# a2 z$ Z- D
32. test ratio_ln
    ( D; s$ F4 d( q1 I+ v/ E2 M5 U
33.         *** Asymptotically equal to ratio1
    . Y2 V: l5 M& w' {! }# _
34.         *** NOTE: This syntax will work in Stata v11
    & t9 p3 p3 s" Y) j
35. 
    
36. ********************
    5 p4 t+ o6 J( d0 a
37. *** FOR STATA 11 ***
    
38. logit bmi3 i.inactive age18 female white diet bmi2, or
    + F4 g& j: ]# n, Y1 v+ S% _
39. margins inactive, post
    ; l( z+ |. r/ l+ h0 N
40. margins, coeflegend
    
41. nlcom (ratio_v11: _b[1.inactive]/_b[0bn.inactive] - 1), post       
    
42. test ratio_v11
    
43.         *** Identical to ratio1
    ; w( U- \2 S+ y/ w8 g
44.         *** Avoids error in Stata v11 by comparing ratio to 0 instead of 1
    
45. 
    
46. 
    - t2 |/ R1 h4 |+ n% N3 e
47. *************************************************************************
    7 F0 h" {. p) @( i& G; s5 O: Q4 \
48. *** Method 1 – Marginal standardization (standard = EXPOSED population)
    ( l- Q8 O$ R3 K# z6 A- ?
49. 
    % m7 u3 o! @* @" L/ ~9 U$ }3 X3 n
50. * First run logistic regression
    * g9 e+ [. G7 O3 M
51. logit bmi3 i.inactive age18 female white diet bmi2, or
    
52. 
    : v$ _! C5 f$ j
53. * Estimate prevalence difference for inactive = 1 vs. inactive = 0
    # v& ]* `, A$ h3 O: \4 T
54. margins if inactive==1, dydx(inactive)
    
55. 
    0 o* A0 c2 t4 Q" [3 c
56. * Estimate predicted probabilities by exposure level
    
57. margins inactive, subpop(inactive) post
    ; S7 B/ \- b" f9 x
58.         *** subpop option calculates weights based on inactive = 1 (exposed group)
    6 R, z5 A( j# O
59. 
    3 ~) C* g0 O- f: k' Z7 n
60. ** Estimate prevalence ratio for inactive = 1 vs. inactive = 0
    , O" n/ A% V: N$ Q6 w! ^' `6 W
61. ***************************
    ) r# H. y2 \+ e0 k3 ~  G9 j
62. *** FOR STATA 12 AND 13 ***
    3 z6 Z' T4 |+ ^+ ]$ X* t
63. margins, coeflegend
    / ~; s% H6 v1 z: O1 F
64. nlcom (ratio1: _b[1.inactive]/_b[0bn.inactive]), post               
    . G4 }; D+ G; B; G. O6 q
65. test ratio1 = 1                                                               
    5 |3 B! E+ R7 D1 M" h, S/ s
66. 
    * N0 ?# r, c( I
67. * Alternative syntax
    
68. logit bmi3 i.inactive age18 female white diet bmi2, or
    
69. margins inactive, subpop(inactive) post
    
70. margins, coeflegend
    
71. nlcom (ratio_ln: ln(_b[1.inactive]) - ln(_b[0bn.inactive] )), post
    - }  b1 ~4 n  V& ]8 l
72. test ratio_ln
    " W, s! z; x% T- a- s, h
73.         *** Asymptotically equal to ratio1
    
74.         *** NOTE: This syntax will work in Stata v11
      w7 H' s6 Q# f* U" m! b
75. 
    
76. ********************
    4 n. T  B% [. Y- z
77. *** FOR STATA 11 ***
    ' H1 F; i' P% N4 h
78. logit bmi3 i.inactive age18 female white diet bmi2, or
    
79. margins inactive, subpop(inactive) post
    ) Z' o- e9 B2 P3 J: ~! y
80. margins, coeflegend
    # A7 c. h* g2 o0 s
81. nlcom (ratio_v11: _b[1.inactive]/_b[0bn.inactive] - 1), post       
    " y% ~$ [9 D0 j% v* _. n7 k. P
82. test ratio_v11
    
83.         *** Identical to ratio1
    * k$ U- N' j8 g7 q6 m4 w* c
84.         *** Avoids error in Stata v11 by comparing ratio to 0 instead of 1
    ( ?% O9 d- }9 h6 O
85. 
    ; S: b" J3 G' C; i
86. ******************************************

复制代码

% N) {* D+ Y. X$ j: P
- Q& K* B- W, }4 F) [
2 _. [8 _* r5 Z/ E