y implemented in only to mention e reading and not alth technology elines
c reviews on the u ected. No good q reviews is presen found an increa increase ranging policy: double re tween 38 and 14
linical
puter-tional were
n the n the t as a
use of quality ted in ase in
from ading 49 per
KCE Reports 17
10,000 women consensus dec women screene specificity with or consensus. T from the recall p In addition, the single-view m mammograms The review of D of double read The consensu mammograms unnecessary as The review of results of Dinne and a decrease or consensus, combined with 3.2.1.2. Com After critical a mammography 2008 40) were m In the review o were calculated main focus is p rate. Next to th variables group The pooled se women) was 86 CI 88.1-88.3%
sensitivity (I2=8 intervals were n 72
n screened), whe creased the recall ed). We see a si unilateral recall an The sensitivity inc policy.
e cancer detection ammograms com (4.4-6.9 per 10,00 Dinnes et al conc ing with arbitratio s or arbitration can decrease ssessment.
Taylor et al, 200 es et al39: an ove e in recall rate afte in contrast to the unilateral recall.
mputer-aided de appraisal of the (CAD), Two revi maintained.
of Noble et al41 7 d where possible put on sensitivity, s hese main variab ped per subgroups ensitivity in the re
6.0% (95%CI 84.
%). Despite the 87.2%) and spec narrow and the se
ereas double rea rate (by between milar pattern for nd an increase in creased with dou n rate increased mpared to the 00 versus 3.0-4.4 cludes that a scre on or consensus
procedure after the number o 0840,40, based on erall increase in t er double reading increased recall etection (CAD) e reviews on c
ews (Noble et al, 7 studies were inc e. As mentioned specificity, recall r bles, other result s, etc will also be eview (based on 2-87.6%) and sp
heterogeneity in cificity (I2= 99.7%
ensitivity analysis w
ading with arbitr n 61 and 269 pe specificity: a dec specificity with ar uble reading, inde more in the stud studies with t per 10,000).
eening protocol co improves the se double reading of women reca 17 studies, conf the cancer detect
combined with ar rate after double
omputer-aided d , 200841 and Tay cluded and pooled
in the methods rate and cancer d ts, such as biops
presented.
n 3 studies, n= 3 ecificity was 88.2 n the estimation
%), the 95% con was robust.
Breast cancer scre
ration or r 10.000 crease in rbitration ependent dies with two-view onsisting ensitivity.
of the lled for firms the
tion rate rbitration reading
detection lor et al, d results section, detection sy rates,
347 324 2% (95%
of the nfidence
In
eening
n comparison w detection rate wit 95%CI 30-80 wom The additional rec 1090-1290). These
ead mammograp 6.3%) were diagn healthy. Unexplain of these findings.
Next to recall rate an additional of 80 Of these women, healthy and 35.9 These rates are women).
The above-mentio which compare sin statistically signific However an incr heterogeneity bet evidence exists fo eading with arbitr 3.2.2. Primary
cohort st 2011 The available evid published after th Findings of prima 3.2.2.1 and 3.2.2.
3.2.2.1. Single n the studies, pub be found specifica
eading. Next t mammograms, fiv
ith single-read m h CAD was 50 w men).
all rate in healthy e women would n phy only. Of thes nosed with cance ned heterogeneity , biopsy rate was 0 per 100 000 bio 65% (95% CI 52
% (95% CI 24.7 calculated with oned review of Ta
ngle reading with cant increase in reased recall ra tween the studie or the improvemen ation compared to Studies: random tudies and cross dence from the rev he search date o ary studies are d
.2)
e reading versus blished between 2 ally on the comp o the clinical ou ve primary stud
mammography th women per 100 women was 1190 not have been rec
e recalled wome r and 96% (95%
y and lack of robus calculated. Base psies (95% CI 60 2.3-76.0%) under 7-48.9%) were di the data of five aylor et al40, 200 single-reading co
cancer detection te was seen, in es. The authors nt in screening pe o single-reading c mized controlled
s-sectional studi views was update of the most rece discussed in the s double reading 2007 and 2011, no
arison of double utcomes after d ies assess more
he incremental ca 000 women scre 0 per 100 000 (95 called based on s
n, 4.1% (95% C
% CI 93.9-97.3%) stness affect cred ed on the CAD sy 0-110) were perfor
rwent biopsy but iagnosed with ca e studies (n= 51 8 gathered 10 st ombined with CA n rate could be fo
ndependent from conclude that erformance with d combined with CA d trials, prospecti ies published 20 ed with primary st ent systematic re following parag g
o primary studies reading versus s ouble reading o e in detail the dibility stem, rmed.
were ancer.
162 tudies AD No
007-tudies eview.
raphs
could single of the
inter-38
observer variab appendix.
Hofvind et al48, Screening Prog 033 870 sc interpretations Consensus wa positive finding concordant pos 3.5%. Of the to following disc differences betw
• Proportion (24.9% ver
• Mass or d cancers (1
• Proportion 15.7%)(p<
• Lobular ca versus 9.1%
Of the total num dismissed disc number of ca conclude that th a lack of comp mammograms.
Caumo et al49, discordant inte consecutive ser discordant inte independent fro cancers (43 co in the arbitrate negative arbitra would be redir
bility. A summary , 2009, based on gram, found, after creening mamm and 21 928 (2 as sought for di gs. 66.8% of the
sitive findings we otal detected canc cordant interpret
ween discordant a of micro-calcific rsus 17.7%)(p<.00 density with mic
1.1% versus 15.4 of DCIS was hig .001)
ancers were less
%)(p=.035) mber of interval ca
cordant interpreta ancers compared
he disagreements petences of the r
2010 examined rpretations on the ries of 7.660 doub erpretations wer om the judgment ncordant and 6 d d cases (5 posit ation implies one
rected for asses
y of results of the n data from the N
r independent, bli mographies, 544 .1%) positive co iscordant interpre e discordant find ere found negati cers (5611 cance ations. There and concordant ca
ations was highe 01)
ro-calcifications w 4%)(p<.001)
gher in discordan s frequent in di ancers (n=1791), ations, revealing d to negative s s on microcalcifica readers in detecti
the effect of a thi e recall rate. In t ble readings of sc re redirected fo t of the third rea
iscordant cancers ive and 1 negativ missed cancer if ssment. Neverthe
e studies can be Norwegian Breast nded double read 447 (5.3%) dis oncordant interpre etations and con dings and 17.9%
ive. The recall ra rs), 23.6% were d were some sig ancers:
er in discordant was lower in dis nt cancers (23.9%
scordant cancers 117 (6.5%) were g a substantially screenings. The
ations are possibl ing microcalcifica ird reader at arbit this study consist creening examina or further asse ader. Of the 49 d
s) 6 cancers are d ve arbitrations). T f only positive arb eless the missed
Breast cancer scre
found in t Cancer ding of 1 scordant etations.
ncordant
% of the ate was detected gnificant cancers scordant
% versus s (7.3%
found in y higher
authors ly due to tions on tration of ting of a tions, all essment, detected detected The one bitrations d cancer
eening
0.13% absolute o arbitration of disco esulting in a 2.8%
overall recall rate consisted of con arbitration of disc preferred practice Ciatto et al50, 2005 ate from 3.82% t women screened d The review of d possible alternativ 1335 cases (1.04 were redirected fo he total of 968 ca al, 200948, the hig
n the group of d eview: sensitivity 99%. Comparing different recall po discordant finding with the approa calcifications: incr value and only a s nter-observer var set-ups were com adiologists, doub adiologist, double wo radiologists, d eferral of all posit he annexes, main ncrease in sensiti had its impact o sensitivity is obt
nterpretation of 4 ncreasing the reca
or 2.0% relative r ordant interpretati
% absolute or 40.9 e in this study w ncordant interpret
cordant interpreta in order to reduc 5 confirms these to 2.59% and the decreased from 4 iscordant interpr ve to arbitration.
%) were reviewe or further assessm ancers) were iden hest proportion o discordant finding y 90%, specificity the recall rate olicies (highest
s due to calcifica ach of only rec rease from 98.98 small increase in r
riability was asse mpared: single-rea ble-reading by r e reading by a ra double reading by tive findings. More
n findings of the ivity for the doubl n cancer detecti tained by a pr readers combine all rate (and unfor
reduction of canc ons would spare 9% relative decre was 528 (6.8%) tations and posi ations is stated b ce the amount of findings: arbitratio e number of canc 4.58 to 4.50.
retations by a c In the study of ed by a consensu ment (US, biopsies ntified. Similar to t of patients with ca gs (32%). Outco y 57%, and neg and cancer de reader recall, un ations), the best calling the patie 8% to 99.66% fo recall rate (0.05%)
ssed by Duijm et ading by radiologi
radiographers an adiographer and y a radiologist an e details on each study were a sig e reading groups ion rate and rec rotocol that take ed with a referral
rtunately the relat
KCE Reports
cer detection rate out 216 assessm ease in recall rate of which 312 c itive arbitrations.
by the authors a (unnecessary) re on reduced the re ers detected per consensus panel
Shaw et al51,51, 2 us panel: 606 (45 s), 71 cancers (7.3 the study of Hofvi alcifications were f omes after conse gative predictive
etection rate bet nanimous recall
results were obt ents with disco or negative pred ).
t al52, 2009. Diff sts, double-readin nd single readin
secondary referr nd a radiographe set-up can be fou nificant (7.3% rel s. The different se
call rate. The hi es into account
of all positive find ted cost and burde
s 172 ecalls.
eferral 1000 ordant dictive ferent ng by
KCE Reports 17
patients). The further assesse national screen 3.2.2.2. Sin Most studies a compare the c outcomes afte combined with reading, this implementation (or even the ar for the screeni results show th studies. The su The study of C also a reductio significant high reading (90.0%
increased reca A later study o statistically sign CAD and reduc Gilbert et al, 2 comparisons combined with was defined as more than 10%
statistically diffe per 1000 wom difference 0.50 with double re (p<0.001) in rec A study within 2009 56) also 72
benefits and pos ed before it can b
ing program.
gle/Double read about the perform
clinical outcomes r single-reading CAD results in kind of screen in clinical practic rbitration reader) b ing units. If we l hat it is more co mmary of these s Ciatto et al, 20035 on in specificity.
er number of det
% versus 85.8%
ll rate (CAD 11.4 of Ciatto et al, 20 nificant difference ced specificity, lea
2008 55found in between cancer CAD was equiv s a 95% confiden
% in either direct erence in cancer men screened w
% (c.i.−7.4% to + eading) in contr call rate with CAD
the national scre came to the sa
ssible harms of t be recommended ding versus CAD mance of CAD in s after double r combined with a higher sensitiv ning tool could
ce. The replacem by a software pro look in detail at omplex than that.
studies can be fou
53 shows an impro There is a sl tected cancer in C
%) but this is 4% versus double 00654 , leads to th
in sensitivity betw ading to an excess
his equivalence r detection rate valent to double
nce interval that tion in the rate detection rate (se with CAD versu +6.6%) or 7.06 pe ast to a small D (3.9% compared ening program in ame conclusion:
this scenario hav d for implementat D
n a screening po reading with the
CAD. If single vity compared to
be advantage ment of the double
ogram could be b the primary stud . We included 5 und in appendix.
ovement in sensit ight, but not sta CAD compared to counterbalanced e-reading 7.9%, p he same conclus ween double read s of false-positive
trial (with matc es) that single reading (i.e. equ ruled out a differ
of cancer detect ensitivity of 87.2%
us sensitivity of er 1000 women s
but significant i d to 3.4%).
Australia (Cawso differences in se
Breast cancer scre
ve to be tion in a
opulation clinical
reading o double
ous for e reader
eneficial dies, the
primary tivity but atistically o double
ched-pair -reading ivalence rence of tion): no
% or 7.02 87.7%, screened increase on et al, ensitivity
b
eening
between CAD an esults were very r These results are 200557. This study program in the U CAD compared t 91.5% versus sin sensitivity by 8.2%
difference was sta assessment) were 10.5% with double versus 5.0% with d 3.2.3. Discussi Both reviews (39 ncreases the sen used). Several cou eading procedure advantages of in mage, this extra r eview of Dinnes e eading may be m shortage of radio eading performan but this is not rel handle with disc arbitration or cons detection rate.
The CAD system adiologist (or ima cancers. However he significant inc patient distress, in attitude towards sc The way CAD is putting marks on
nd double readin reader dependent e contradicted by y, done within th K, demonstrates to single reading ngle-reading 90.2
% (sensitivity of 98 atistically significa e significant high e reading for reca double reading fo ion
, 40) come to th nsitivity and decr untries, including e in their nationa ndependent doub reading procedure et al, 200139 only
ore cost-effective ologists in some
nce of radiograph levant for the Be cordant interpret sensus) has its im m is a suppleme
age reader), aim r, this small increa crease in recall ra ncreased number creening mammo
done implies a s suspicious mass
ng was not stati t.
the results of the he framework of t
an increased se g (single-reading
%), but double r 8.4%), they did not ant. Recall rates
er with CAD: 13 ll rate for arbitratio r recall rate for as
he same conclu reases the recall Belgium, have im al screening progr ble reading the
e implies an incre indirect evidence e compared to sing
countries, resea hers to the perfor elgian situation.
tations (recall o mpact on the reca
ntary tool to the ing to increase th ase in sensitivity ate. The increase r of health care v
graphy.
software-matic an ses, followed by
stically significan e study of Khoo the national scre ensitivity by 1.3%
combined with reading increased t formally assess (for arbitration an 3.8% with CAD v
on and 6.1% with ssessment.
usion: double re l rate (if arbitrati mplemented the d ram. Nevertheles same mammogr eased workload. I was found that d gle reading. Due t archers compared rmance of radiolo
The approach ho of all these find all rate and the ca e interpretation o he number of det
is counterbalance ed recall rate lea visits and a chan nalysis of the im reading by a h raphic In the ouble to the d the ogists,
ow to
40
reader and a d are false-positi makes that the specificity, an in of biopsies (of advantage in sc increased rate missed by the f et al, 2008 41 representativen importance of th The interpretati primary studies of blinding to c and the lack o specificity and s the restricted fo missed due to negatives.
Apart from th screening prog interval-screeni Key points
• Double rea
o increa
o decrea
• double rea programs
• Single-rea
o Only a
o a sign
recall
decision on which ives. But the hig e human reader i ncreased time to
healthy women).
creening for brea is determined by first reader. Only 2
reported the typ ness and general
hese detected can on of the results i s in the review, ca clinical information of reporting abou sensitivity of the C ollow-up period o o this limited fol e methodologica gram varies betw
ngs, etc.
ading:
ases the sensitiv ases the recall ra ading is widespr ading combined w
a small increase nificant increase
rate
h marks are true-p gh number of po is overwhelmed, read the images a
An increased ca ast cancer, but the
y the stage and t 2 of the four stud pe and stage of
isability of these ncers is not asses is limited by the p aused by the retro n (or lack of repo ut the case sele CAD system could of one year. Slow
low-up time and al limitations, th ween countries,
vity compared to ate if arbitration read used in natio
with CAD:
in sensitivity of false-positive
positives and whi ositive marks pe resulting in a de and an increased ancer detection ra
e real added valu type of the cance ies in the review o f cancer, this lim
finding. Also the ssed.
poor internal validi ospective design, orting about this b ection methods. A d be overestimate w-growing cancers d this will lead t e set-up of a with differences
single reading is applied onal screening
e marks and incre
Breast cancer scre
ich ones er image ecreased
number e clinical
ity of the the lack blinding) Also the ed due to
s can be to
false-national in age,
eening
CAD enhance the clinical o rate) to doub The high num additional ar enhances the
3.3. Full-field
3.3.1. Systemaassessm After critical ap mammography (F made it difficult to we decided to re-a critical appraisal o 3.3.2. Primary
cohort st selected n the review of esults of the pr screening program was given to the biopsy rate, secon
he population of mammography (F detection in a p nationally organize 3.3.2.1. Lewin The results of the
he publications of n the study of Lew and older were ex SFM and re-exam positive findings (
es the reading p outcomes are com ble reading
mber of false-pos bitration, which e implementation