Supplemental Digital Content 2 – Conversions and Methodology
Mean and standard deviation
Mean is defined as follows:
´ x= ∑
i=1 n
x
in
There are n number of observations (usually patients in health economics applications).
x
i is the attribute of interest (e.g., visual acuity gains of patienti
).Variance is defined as follows:
σ
2= ∑
i=1
n
( x
i− ´ x )
2n
Standard deviation (SD) is
σ = √ σ
2 . See page 81 of Hogg and Tanis [1] for a discussion on mean and variance using empirical distribution.Subgroup means to aggregate mean
Observations belong to k number of mutually exclusive and collectively exhaustive
subgroups
S
1, S
2, … , S
k . The aggregation of subgroup means to aggregate mean uses the following formula:´ x= n
1´ x
1+n
2´ x
2+…+ n
k´ x
kn
j is the number of observations for subgroupn j
and´ x
j is the subgroup mean for subgroupj
.Proof:
n
1´ x
1+ n
2x ´
2+…+ n
kx ´
kn =
n
1∑
i∈S1
x
in
1+n
2∑
i∈S2
x
in
2+ …+ n
k∑
i∈Sk
x
in
kn
¿i
∑
∈S1x
i+ ∑
i∈S2
x
i+ …+ ∑
i∈Sk
x
in
¿
∑
i=1 nx
in
¿
´ x
Subgroup SD to aggregate SD
σ
2= n
1(σ
12+ ´ x
12)+n
2( σ
22+ ´ x
22)+…+n
k(σ
k2+ ´ x
k2)
n −´ x
2Proof:
Note that
∑
i=1 nx
i2n =
∑
i=1 n( x
i−´ x )
2+2 x
ix ´ −´ x
2n
¿∑
i=1 n( x
i−´ x )
2n +
∑
i=1 n2 x
i´ x
n −
∑
i=1 n´ x
2n
¿σ
2+2 ´ x
∑
i=1 nx
in − n ´ x
2n
¿
σ
2+2 ´ x
2−´ x
2 ¿σ
2+ ´ x
2Consequently,
n ( σ
2+ ´ x
2) = n
∑
i=1 nx
i2n
¿
∑
i=1 n
x
i2 ¿∑
i∈S1
x
i2+ ∑
i∈S2
x
i2+ …+ ∑
i∈Sk
x
i2¿
n
1( σ
12+ ´ x
12) +n
2( σ
22+ ´ x
22) + …+ n
k( σ
k2+ ´ x
k2)
Arrange the above expression in terms of σ2 .
σ
2= n
1( σ
12+ ´ x
12) + n
2( σ
22+ ´ x
22) +…+ n
k( σ
2k+ ´ x
k2)
n −´ x
2σ = √ n
1( σ
12+ ´ x
12) + n
2( σ
22+ ´ n x
22) +…+ n
k( σ
2k+ ´ x
k2) − ´ x
2logMAR and Early Treatment Diabetic Retinopathy Study (ETDRS) score
Mean logMAR score to mean ETDRS score
The formula was derived is based on data described in Beck et al. [2], which showed an obvious linear relationship between logMAR score and ETDRS score.
y= − x−1.7 0.02
For example, logMAR value of 0.9 is
−0.9−1.7
0.02
=40 in ETDRS score.SD of logMAR to SD of ETDRS letters
σ
2ETDRS= ( 0.02 −1 )
2σ
logMAR2↔ σ
ETDRS=50 ×σ
logMARProof:
VAR (aX + b)= a
2× VAR ( X)
, whereX
is a random variable anda
andb
are constants. See page 80 of Hogg and Tanis [1] for the derivation.Apply this equation on the linear transformation of logMAR score to ETDRS score:
σ
2ETDRS=VAR ( − x−1.7 0.02 )
¿VAR ( 0.02 −1 × x+ 1.7
0.02 )
¿( 0.02 −1 )
2VAR ( x )
¿( 0.02 −1 )
2σ
logMAR2Take square root on both side of the above equation.
σETDRS=50× σlogMAR
Change in mean of logMAR score to change in mean of ETDRS score y= − x
0.02
Taking difference at the beginning and the end of observation period deletes the constant.
y=ybeginning−yend ¿−
x
beginning−1.7
0.02 + x
end−1.7
0.02
¿−x
beginning− x
end0.02
¿−x
0.02
The variance of the change in mean logMAR score can be calculated using the same formula for the variance of mean logMAR score because deleting the constant (
−1.7
0.02
¿ has no effect on variance (VAR ( aX + b)=VAR(aX )
).Decimals to ETDRS letters y=
ln ( x 0.0199 ) 0.0461
The above formula was derived based on the conversion table in Elliott [3].
Standard error to standard deviation σ
2=n× s . e .
2↔ σ = √ n ×s . e .
2See Barde and Barde [4] for the definition.
95% confidence interval to standard error ( upper limit− lowerlimit
s . e .=¿ ¿ 3.92
See section 7.7.7.2 of Higgins and Green [5].
Higgins and Green [5] did not provide a justification for the above equation. According to the definition of confidence interval, the following must hold.
P
(
lower limit ≤ μ ≤upper limit)
=95 %Confidence interval is typically estimated as follows:
P ( ´ x −1.96 × s . e .≤ μ ≤ x ´ +1.96 × s. e . ) ≈ 95 % ,
Consequently,
upper limit −lower limit=( ´ x + 1.96× s . e . )−( ´ x−1.96 × s . e .)
↔upper limit−lower limit=
(
x− ´´ x)
+(
1.96× s . e .+1.96× s .e .)
↔upper limit−lower limit=3.92× s . e . (upper limit−lowerlimit
↔ s . e .=¿ ¿ 3.92
The p-value to the t-value
t=tinv
(
p , df)
,tinv
: the inverse of the t-distribution df : degrees of freedomp
: the p-valueSee section 7.7.3.3 of Higgins and Green [5]. This is a straightforward application of an inverse function.
Coefficient estimate and t-value to standard error s . e .= ^ β
t ,
^ β
: coefficient estimatet
: the t-valueSee section 7.7.3.3 of Higgins and Green [5].
Again, Higgins and Green [5] did not provide a justification for the above equation. The definition of the t-value is as follows.
t = β ^ − β s . e .
β
is a population parameter, which is not random. Typically, null hypothesis isβ=0
. For example, one can be interested whether the mean difference in visual acuity at baseline and atyear 2 is zero or not. In this case, the expression is reduced to
t = ^ β
s . e .
. A simple rearrangement leads tos . e .= ^ β
t
.References
1. Hogg RV, Tanis EA. Probability and statistical inference. 7th ed. Pearson. 2005.
2. Beck RW, Moke PS, Turpin AH, et al. A computerized method of visual acuity testing:
adaptation of the early treatment of diabetic retinopathy study testing protocol. Am J Ophthalmol.
2003;135(2):194–205. doi:10.1016/S0002-9394(02)01825-1
3. Elliott DB. The good (logMAR), the bad (Snellen) and the ugly (BCVA, number of letters read) of visual acuity measurement. Ophthalmic Physiol Opt. 2016;36(4):355–58.
doi:10.1111/opo.12310
4. Barde MP, Barde PJ. What to use to express the variability of data: standard deviation or standard error of mean? Perspect Clin Res. 2012;3(3):113–16. doi:10.4103/2229-3485.100662 5. Higgins JP, Green S. Cochrane handbook for systematic reviews of interventions. Vol 4.
Chichester, UK: John Wiley & Sons; 2011.