{smcl}
{* 1 Sept 2005}{...}
{hline}
help for {hi:binscrn3} {right: (version 1.1.0) }
{hline}

{title:Comparison of binary screening tests - paired data}

{p 8 12}{cmd:binscrn3}
{it:test1_var}
{it:test2_var}
{it:disease_var}
{ifin}
[, {opt l:evel(#)}
{opt f:ormat(%fmt)}
{opt ni:ter(#)}]


{title:Description}

{pstd}
{cmd:binscrn3} compares two screening tests with respect to true and false positive
rates, positive and negative predictive values, and diagnostic likelihood
ratios for binary test result data.  Comparisons assume paired data - that
test results for both tests are observed for each subject.

{pstd}
Positive and negative predictive values assume cohort data; i.e. that the proportion of
disease positive observations in the dataset represents the population disease
prevalence.

{pstd}
{it:test1_var} and {it:test2_var} are 0/1 test result variables and
{it:disease_var} is the  0/1 disease status indicator variable.  Data are
assumed to be organized in wide format, i.e. with a single observation per
subject.

{pstd}Relative screening measures and confidence limits are available as returned results after running the
program:

{col 8} rTPF:  r(rtpf)   [r(rtpf_lb), r(rtpf_ub) ]
{col 8} rFPF:  r(rfpf)   [r(rfpf_lb), r(rfpf_ub) ]
{col 8} rPPV:  r(rppv)   [r(rppv_lb), r(rppv_ub] ]
{col 8} rNPV:  r(rppv)   [r(rppv_lb), r(rppv_ub) ]
{col 8}rDLR+:  r(rdlrp)
{col 8}rDLR-:  r(rdlrn)

{pstd}along with the following joint rectangular confidence regions:

{col 8}joint rFPF,rTPF: r(fpfj_lb) r(fpfj_ub) x r(tpfj_lb) r(tpfj_ub)


{title:Options}

{phang}
{opt level(#)}; see {help estimation options##level():estimation options}.

{phang}
{opt format(%fmt)} specifies the display {help format} for estimates and CI's.  The
    default is %5.3g

{phang}
{opt niter(#)} specifies the maximum number of iterations allowed (see {help maximize}) for
fitting the {help glm:GLM} regression model used to obtain rPPV and rNPV confidence bounds.  If convergence is not
achieved, another attempt using the {it:difficult} maximization option is made
under the same iteration limit.  The default is 20.


{title:Remarks}

{pstd}
Confidence limits and one of the hypothesis test options for rTPF and rFPF are based
on the large sample normal distribution of the log measures
per Cheng and Malcuso (see References).

{pstd} Rectangular joint confidence regions (rFPF,rTPF) are based on
on sqrt(1-alpha) level confidence limits for the component ratios
rFPF, rTPF.

{pstd} Confidence limits for the rPPV and rNPV are based on a marginal
probability GLM agreement model (Pepe, section 3.6.2) which is fit with
separate observations for each test result and subject, using a sandwich estimator of
the variance which clusters on subject.

{pstd} See {help binscrn1} to obtain various screening measures and confidence limits
for results from a single screening test with binary outcome and {help binscrn2}
for comparison of tests with unpaired data.


{title:References}

{pmore}Pepe MS.  {browse "http://www.fhcrc.org/science/labs/pepe/book":The Statistical Evaluation of Medical Tests for Classification and Prediction.}
Oxford University Press.  Chapter 3.  2003.
{break}

{pstd}calculation of confidence intervals for relative TPF and FPF measures per:

{pmore}Cheng H, and Macaluso M.  Comparison of the accuracy of two tests with a
confirmatory procedure limited to positive reuults.  Epidemiology 8:104-6, 1997.


{title:Examples}

{phang} .use http://www.fhcrc.org/science/labs/pepe/book/data/est1, clear {p_end}
{phang} .binscrn3 y2 y1 d {p_end}
{phang} .binscrn3 y2 y1 d, level(90) format(%6.2g){p_end}


{title:Author}

{pstd} Gary Longton, Fred Hutchinson Cancer Research Center, Seattle, WA.{break}
glongton@fhcrc.org