Factors associated with attrition in a longitudinal online study: results from the HaBIDS panel.

RESEARCH ARTICLE Open Access Fac o s associa ed wi h a i ion in a longi udinal online s udy: esul s om he HaBIDS panel Nicole Rübsamen 1,2 , Manas K. Akma o 1,3,4 , S e anie Cas ell 1,3 , And é Ka ch 1,2 and Ra ael T. Mikolajczyk 1,5,6* Abs ac Backg ound: Knowing abou p edic o s o a i ion in a panel is impo an o ini ia e ea ly measu es agains loss o pa icipan s. We in es iga ed a i ion in bo h ea ly and la e phase o an online panel wi h special ocus on p e e ences ega ding mode o pa icipa ion. Me hods: We used da a om he HaBIDS panel ha was designed o in es iga e knowledge, a i udes, and p ac ice ega ding in ec ions in he Ge man gene al popula ion. HaBIDS was di ided in o wo phases: an ini ial phase when some pa icipan s could choose hei p e e ed mode o pa icipa ion (pape -and-pencil o online) and an ex ended phase when pa icipan s we e asked o become membe s o an online panel ha was no limi ed ega ding i s du a ion (i.e. pa icipan s ini ially p e e ing pape ques ionnai es swi ched o online pa icipa ion). Using compe ing isks eg ession, we in es iga ed wo ypes o a i ion ( o mal wi hd awal and discon inua ion wi hou wi hd awal) among online pa icipan s, sepa a ely o bo h phases. As po en ial p edic o s o a i ion, we conside ed sociodemog aphic cha ac e is ics, physical and men al heal h as well as auxilia y in o ma ion desc ibing he su ey p ocess, and, in he ex ended phase, ini ial mode p e e ence. Resul s: In he ini ial phase, highe age and less equen In e ne usage p edic ed wi hd awal, while younge age, highe s ess le els, delay in e u ning he consen o m, and need o ecei ing eminde emails p edic ed discon inua ion. In he ex ended phase, only need o ecei ing eminde emails p edic ed discon inua ion. Numbe s o wi hd awal in he ex ended phase we e oo small o analysis. Ini ial mode p e e ence did no p edic a i ion in he ex ended phase. Besides age, he e was no e idence o di e en ial a i ion by sociodemog aphic ac o s in any phase. Conclusions: P edic o s o a i ion we e simila in bo h phases o he panel, bu hey di e ed by ype o a i ion (wi hd awal s. discon inua ion). Sociodemog aphic cha ac e is ics only played a mino ole o bo h ypes o a i ion. Need o ecei ing a eminde was he s onges p edic o o discon inua ion in any phase, bu no p edic o o wi hd awal. We ound p edic o s o a i ion, which can be iden i ied al eady in he ea ly phase o a panel so ha coun e measu es (e.g. special incen i es) can be aken. Keywo ds: A i ion, Heal h su ey, In e ne , Longi udinal s udy, Mixed-mode, Online, Panel, Pa icipa ion, Response, Wi hd awal * Co espondence: [email p o ec ed] 1 Depa men o Epidemiology, Helmhol z Cen e o In ec ion Resea ch (HZI), Inho ens aße 7, 38124 B aunschweig, Ge many 5 Hanno e Medical School, Hanno e , Ge many Full lis o au ho in o ma ion is a ailable a he end o he a icle © The Au ho (s). 2017 Open Access This a icle is dis ibu ed unde he e ms o he C ea i e Commons A ibu ion 4.0 In e na ional License (h p://c ea i ecommons.o g/licenses/by/4.0/), which pe mi s un es ic ed use, dis ibu ion, and ep oduc ion in any medium, p o ided you gi e app op ia e c edi o he o iginal au ho (s) and he sou ce, p o ide a link o he C ea i e Commons license, and indica e i changes we e made. The C ea i e Commons Public Domain Dedica ion wai e (h p://c ea i ecommons.o g/publicdomain/ze o/1.0/) applies o he da a made a ailable in his a icle, unless o he wise s a ed. Rübsamen e al. BMC Medical Resea ch Me hodology (2017) 17:132 DOI 10.1186/s12874-017-0408-3 Backg ound In 2005, Eysenbach a gued ha a “science o a i ion” is needed [1]. He dis inguished wo ypes o a i ion in he con ex o in e en ional eHeal h applica ions: use s who o mally wi hd aw and use s who do no o mally wi hd aw, bu who a e no longe using he applica ion (discon inua ion). Eysenbach s a es ha hese ypes o a i ion can ei he be closely ela ed o no , i.e. ha i is possible o ha e low a es o wi h- d awal, bu s ill many pa icipan s no using he appli- ca ion (he ci es [2, 3] as examples). He hypo hesized ha a high p opo ion o wi hd awal is a esul o dis- con inua ion. Following Eysenbach, se e al au ho s ha e in es iga ed he ypes o a i ion in eHeal h in- e en ions [4–7]. Howe e , since eHeal h in e en- ions a e mos ly ailo ed o he indi idual pa icipan , hei indings abou a i ion migh no be ans e able o online panels (which do no include indi idualized in e en ions). In online panels, he p oblem o a i ion was s udied in he con ex o social and poli ical sciences [8–10], showing ha sociodemog aphic ac o s as well as o he ac o s possibly changing o e ime, e.g. commi men o he su ey o panel a igue, we e associa ed wi h a - i ion [8]. Da a on a i ion in heal h- ela ed online panels is es ic ed o speci ic esea ch ields, e.g. syn- d omic su eillance [11] o changes in heal h among adolescen s [12]. I is no clea i p edic o s o a i ion in heal h- ela ed online panels a e compa able o hose in panels in he con ex o social sciences. Knowing who is a isk o a i ion in a panel is impo an o make e o s o keep hose pa icipan s a isk engaged in hepaneland obeable op edic biasaspanela - i ion can limi gene alizabili y and make analyses o epea ed ( ollow-up) measu emen s mo e complica ed and less alid [7]. Wi h inc easing use o he in e ne , s udies using online da a collec ion, especially wi h mobile applica- ions, will likely become mo e common in u u e; s ill i is no clea i hose who would –gi en he choice –p e e o he modes o pa icipa ion display dis inc pa e ns o a i ion when pa icipa ing in an online panel. We ook ad an age o da a om a longi udinal panel on heal h- ela ed issues and in es iga ed wo ques ions: 1) which ac o s a e associa ed wi h su ey a i ion, and 2) i a i ion di e s be ween hose who ini ially pa ici- pa ed in he panel by pape -and-pencil and ag eed o swi ch o online pa icipa ion la e and hose who pa ic- ipa ed online om he beginning. We di e en ia ed be ween wi hd awal and discon inua ion because we hy- po hesized ha hese wo en i ies ep esen di e en g oups o pa icipan s ha ha e di e en easons o s op pa icipa ing in he s udy. Me hods Rec ui men This analysis was based on he popula ion-based, longi- udinal Hygiene and Beha iou In ec ious Diseases S udy (HaBIDS), conduc ed in ou egions o Lowe Saxony, Ge many and designed o assess knowledge, a i udes, and p ac ice ela ed o in ec ious diseases and o in es i- ga e e ec s o su ey design [13, 14]. Po en ial pa ici- pan s be ween 15 and 69 yea s o age we e d awn by means o p opo ional s a i ied andom sampling om he egional popula ion egis ies. We sampled indi id- uals p opo ionally om 22 age-sex s a a (11 age g oups mul iplied by wo sexes). We did no explici ly aim o include o exclude immig an s, bu only included he ac ion o immig an s ha li ed in he s udy egions a he ime o sampling. Pa icipan s in he s udy we e asked i hey we e bo n in Ge many. We did no include a ques ion abou i s o second deg ee immig a ion. In a i s wa e (Janua y 2014), we used a mixed-mode app oach, i.e. we o e ed a choice be ween pape -and- pencil and online pa icipa ion o 16,895 po en ial pa - icipan s li ing in he egions B aunschweig and Vech a. We e e o indi iduals who chose pape -and-pencil pa - icipa ion as “mixed-mode: pape ”g oup and o indi id- uals who chose online pa icipa ion as “mixed-mode: online”g oup (Table 1 gi es an o e iew o he g oups de ined h oughou his a icle). In a second wa e (Ap il 2014), we o e ed only online pa icipa ion o 10,000 newly in i ed po en ial pa icipan s li ing in Salzgi e and Wol enbü el. We e e o pa icipan s om his second wa e as “online-only”g oup. All po en ial pa ic- ipan s ecei ed one in i a ion le e ia land mail and no u he eminde le e s. To s imula e con inuing pa ici- pa ion, we o e ed bimon hly eedbacks abou esul s o he s udy. S udy phases We s a ed in he in i a ion le e o he HaBIDS s udy ha he s udy would consis o nine online o wo pape -and-pencil ques ionnai es, i.e. ha he s udy would be limi ed ega ding i s du a ion. Be ween Janua y 2014 and July 30 h, 2015 (“ini ial phase”), online pa ici- pan s, i.e. pa icipan s in “mixed-mode: online”and “on- line-only”, ecei ed ques ionnai es A o K (Addi ional ile 1 illus a es he iming, leng h, opics, and esponse a es o all ques ionnai es; Addi ional ile 2 p o ides he English ansla ion o all ques ionnai es). The “online- only”pa icipan s ecei ed ques ionnai es A and B a he same ime o compensa e o hei delayed ec ui - men in o he panel. The “mixed-mode: pape ”pa ici- pan s ecei ed wo pape ques ionnai es: he i s one included he opics A, B, C, D and he second one in- cluded E, G, H, I, K. Ques ionnai es F and J we e sen o online pa icipan s only because hey we e added in he Rübsamen e al. BMC Medical Resea ch Me hodology (2017) 17:132 Page 2 o 11 cou se o he s udy o su ey pa icipan s abou ecen opics (Ebola, In luenza). In an ex ended phase s a ing on July 30 h, 2015, we in i ed all pa icipan s o he ini ial phase (who had no o mally wi hd awn so a ) o con inue wi h he s udy and become pa o he HaBIDS online panel ha was no limi ed ega ding i s du a ion. In i a ion o his ex- ended phase s a ed ha new ques ionnai es would be sen o panel membe s e e y wo o h ee mon hs. Fo “mixed-mode: pape ”pa icipan s, becoming pa o his online panel mean o swi ch he mode o pa icipa ion. I a pa icipan did no wan o become pa o he HaBIDS online panel, we asked abou he easons o his decision in a mul iple choice ques ion. All panel membe s ecei ed i e ques ionnai es (ques ionnai e L o P, Addi ional ile 1) be ween Augus 2015 and May 2016. Fo each online ques ionnai e (A o P), we sen a sin- gle eminde (email) o pa icipan s who had no illed in he ques ionnai e wi hin wo weeks a e he ini ial in- i a ion. In each email ha he pa icipan s ecei ed (in- i a ions o ques ionnai es as well as eminde emails), pa icipan s we e old explici ly ha i hey wish, hey can wi hd aw om he s udy and u he eminde s by eplying o he email. A low diag am ha summa izes ec ui men p ocess and s udy phases is p esen ed in Addi ional ile 3. Resea ch ques ions Figu e 1 p esen s an o e iew o ou esea ch ques ions, on which we elabo a e below. Ou i s aim was o in es iga e which ac o s a e asso- cia ed wi h su ey a i ion. We we e especially in e - es ed in in es iga ing i he same a iables we e associa ed wi h a i ion in bo h he ea ly and he la e phase o an online panel. I hey would di e , hen i would be ad isable o ailo coun e measu es, e.g. special incen i es, o he di e en phases o an online panel. We in es iga ed (1a) which ac o s a e associa ed wi h a i- ion in he ini ial phase o he online panel and (1b) which ac o s a e associa ed wi h a i ion in he ex- ended phase o he online panel. Ou second aim was o in es iga e i a i ion di e s be ween hose who ini ially pa icipa ed in he s udy by pape -and-pencil and ag eed o swi ch o online pa ici- pa ion la e , and hose who pa icipa ed online om he beginning. To assess i he e is selec ion bias in he ex- ended phase, we in es iga ed i pa icipan s wi h ce ain sociodemog aphic cha ac e is ics we e less likely o be ans e ed om he ini ial s udy o he online panel. We in es iga ed (2a) which ini ial pape -and-pencil pa - icipan s ag ee o con inue wi h he online panel, (2b) which ini ial online pa icipan s ake pa in he ex- ended phase and (2c) i a i ion in he online panel di - e s be ween hose who ini ially op ed o pape -and- pencil ques ionnai es compa ed o hose who op ed o online pa icipa ion. De ini ion o a i ion We di ided he online pa icipan s in he ini ial phase as well as he online panel membe s in he ex ended phase in o h ee mu ually exclusi e ou come g oups: wi h- d awe s ( hose who o mally e oked hei pa icipa ion), discon inue s ( hose who did no ill in a leas wo con- secu i e ques ionnai es and emained non- esponde ; see o Addi ional ile 4 ha illus a es discon inua ion), and egula use s (pa icipan s who nei he wi hd ew no dis- con inued pa icipa ion). Acco ding o he me hod p o- posed by Eysenbach [1], we es ic ed he in es iga ed g oups o hose pa icipan s who had illed in he i s ques ionnai e o he co esponding phase, i.e. ques ion- nai e A o AB in he ini ial phase and ques ionnai e L in he ex ended phase, espec i ely. Table 1 De ini ion o exp essions used h oughou he a icle Exp ession De ini ion Phase 1: Ini ial phase S udy phase be ween ini ial in i a ion o he HaBIDS panel and in i a ion o he imely unlimi ed panel (includes ques ionnai es A o K) “Mixed-mode: pape ”g oup g oup o indi iduals who chose o pa icipa e ia pape -and-pencil ques ionnai es in he mixed-mode su ey “Mixed-mode: online”g oup g oup o indi iduals who chose o pa icipa e ia online ques ionnai es in he mixed-mode su ey “Online-only” g oup g oup o indi iduals who pa icipa e in he online-only su ey (pa icipa ion only possible ia online ques ionnai es) Ini ial online pa icipan s g oup o indi iduals who pa icipa e ia online ques ionnai es in bo h su eys, i.e. union o he wo g oups “mixed-mode: online”and “online-only” Phase 2: Ex ended phase (online panel) S udy phase be ween in i a ion o he imely unlimi ed panel and la es ques ionnai e (includes ques ionnai es L o P) Fo me pape pa icipan s “mixed-mode: pape ”who con inued pa icipa ing in he ex ended phase, i.e. who swi ched om pa icipa ion ia pape -and-pencil o online ques ionnai es All- ime online pa icipan s online pa icipan s who con inued pa icipa ing in he ex ended phase Bo h phases Wi hd awe s pa icipan s who o mally wi hd aw om HaBIDS Discon inue s pa icipan s who do no o mally wi hd aw, bu who a e no longe illing in ques ionnai es ( his condi ion is ul illed i a leas wo consecu i e ques ionnai es a e missing and he pa icipan does no e u n o he s udy; see o Addi ional ile 4 ha illus a es discon inua ion) Regula use s pa icipan s who nei he wi hd aw no discon inue Rübsamen e al. BMC Medical Resea ch Me hodology (2017) 17:132 Page 3 o 11 The da e o wi hd awal was de ined as he da e o in i a ion o he ques ionnai e in ela ion o which wi hd awal was decla ed. The da e o discon inua ion was de ined as he da e o in i a ion o he i s ques- ionnai e ha was no illed in. In ew cases, email deli e y ailed a some poin du ing he s udy so ha pa icipan s did no longe ecei e in i a ion emails. In hese cases, he pa icipan was excluded comple ely om he analyses o he espec i e phase, i.e. ini ial o ex ended phase. P edic o s o a i ion As po en ial p edic o s o a i ion, we conside ed socio- demog aphic cha ac e is ics, sel - epo ed physical and men al heal h as well as me ada a (auxilia y in o ma ion desc ibing he su ey p ocess). Sociodemog aphic cha - ac e is ics comp ised age, sex, ma i al s a us (ma ied, unma ied, di o ced, widowed), highes comple ed edu- ca ional le el (lowe seconda y educa ion o app en ice- ship, s ill a uppe seconda y school, uni e si y en ance quali ica ion [ h ough uppe seconda y educa ion o o- ca ional school], and uni e si y deg ee), and equency o In e ne usage (daily and less han daily). Da a abou physical and men al heal h included sel - a ed heal h s a us (excellen , e y good, good, ai , poo ) [15], WHO-5 well-being index (sum o i e i ems, sco e anges om 0 [poo well-being] o 100 [excellen well- being]) [16], pe cei ed s ess scale (PSS, mean o ou S ill pa icipa ing a he ime o analysis Pa icpa ing in he ex ended phase S ill pa icipa ing a he end o he ini ial phase Pa icipa ing in he ini ial phase A i ion in he in ial phase Swi ching modes A i ion in he ex ended phase (2a) Which ini ial pape -and-pencil pa icipan s ag ee o con inue wi h he online panel? (2b) Which ini ial online pa icipan s ake pa in he ex ended phase? (2c) Does a i ion in he online panel di e be ween hose who ini ially op ed o pape -and-pencil ques ionnai es compa ed o hose who op ed o online pa icipa ion? (1b) Which ac o s a e associa ed wi h a i ion in he ex ended phase o he online panel? Ini ial phase (ques ionnai es A o K, Janua y 2014 o July 2015) In i a ion o con inue wi h he panel (pape -and-pencil pa icipan s swi ch o online mode) Ex ended phase (ques ionnai es L o P, Augus 2015 o May 2016) (1a) Which ac o s a e associa ed wi h a i ion in he ini ial phase o he online panel? Aims 1. To in es iga e which ac o s a e associa ed wi h su ey a i ion 2. To in es iga e i a i ion di e s be ween hose who ini ially pa icipa ed in he panel by pape -and-pencil and ag eed o swi ch o online pa icipa ion la e and hose who pa icipa ed online om he beginning Fig. 1 O e iew o he esea ch ques ions Rübsamen e al. BMC Medical Resea ch Me hodology (2017) 17:132 Page 4 o 11 i ems, sco e anges om 0 [ e y low s ess le el] o 16 [ e y high s ess le el]) [17], and mean sco e based on in ec ions and in ec ion-associa ed symp oms in he las 12 mon hs (“ID Sc een”, sco e anges om 0 [no in ec- ions in he las 12 mon hs] o 46 [mo e han 42 in ec- ions in he las 12 mon hs]) [18]. Me ada a included ini ially p e e ed mode o pa ici- pa ion (online e sus pape -and-pencil), ime be ween in i a ion and e u n o he signed in o med consen o m (in days), and whe he he pa icipan esponded only a e ecei ing a eminde email o he i s online ques ionnai e (i.e. ques ionnai e A in he ini ial phase and ques ionnai e L in he ex ended phase). We only looked a he need o eminde s o he i s ques ion- nai e o in es iga e i al eady a his ime poin la e a - i ion can be p edic ed. In he analysis o po en ial sel -selec ion in o he ex- ended phase (ques ions 2a and 2b), we also conside ed numbe o ques ionnai es illed in du ing he ini ial phase and o al numbe o eminde s ecei ed du ing he ini ial phase. S a is ical analysis We compa ed he s udy popula ion’s composi ion wi h he a ge popula ion’s (inhabi an s o Lowe Saxony be- ween 15 and 69 yea s o age [19]) composi ion by di id- ing each popula ion in s a a o age, sex, and educa ion and calcula ing he ac ion o each s a um in he e- spec i e popula ion o assess he possibili y o gene a ing gene alizable es ima es ia pos -s a i ica ion. To answe ques ion 1a, we analysed da a om he ini- ial phase in h ee s eps. To in es iga e how p edic o s we e associa ed wi h each ype o a i ion, we build sep- a a e Cox p opo ional haza ds eg ession models o ob ain unadjus ed haza d a ios (sepa a ely o discon- inua ion and wi hd awal) [20]. To make conclusions on cumula i e isks, we hen applied compe ing isks e- g ession [21] in combina ion wi h he leas absolu e sh inkage and selec ion ope a o me hod (LASSO) o simul aneously selec p edic o a iables and es ima e eg ession pa ame e s [22]. We buil wo models (R package “c p”) o in es iga e i p edic o s di e ed by ype o a i ion: model A included discon inua ion as e en o in e es and wi hd awal as compe ing e en while model B included wi hd awal as e en o in e es and discon inua ion as compe ing e en . As uning pa - ame e , we used he λ ha minimized BIC (as desc ibed in [22]). We compa ed models A and B ega ding he se- lec ed a iables. Finally, i a p edic o a iable was se- lec ed in bo h model A and B, bu he associa ions had opposi e di ec ions o he wo ypes o a i ion, we used a nonpa ame ic es o end ac oss o de ed g oups [23] o in es iga e he o e all associa ion be ween his p edic o a iable and he composi e endpoin (a - i ion because o wi hd awal o discon inua ion). To answe ques ion 2a and 2b, we applied he LASSO in logis ic eg ession (R package “glmne ”, dependen a iable = pa icipa ion in he ex ended phase) o he subse o pape -and-pencil o online pa icipan s, e- spec i ely (models C and D). Fo answe ques ions 1b and 2c, we applied he LASSO o da a om he ex ended phase. As only ew pa icipan s wi hd ew du ing his phase, we used he LASSO in Cox eg ession (R package “glmne ”) wi h dis- con inua ion as dependen a iable (model E) and ex- cluded pa icipan s who had wi hd awn. To accoun o he e ec ha all- ime online pa icipan s we e al eady used o online ques ionnai es, we used he χ 2 es o compa e discon inua ion among he o me pape -and- pencil pa icipan s be ween ques ionnai es L and P and among he “online-only”pa icipan s (who we e also “ o ced” o use he online mode) be ween ques ionnai es A and E. We conduc ed comple e-case analyses and conside ed p≤0.05 as signi ican . To check he p opo ional haz- a ds assump ion, we plo ed Schoen eld esiduals p io o Cox and compe ing isks eg essions. To assess pos- sible nonlinea i y o con inuous p edic o s, we es ima ed ac ional polynomials p io o all eg ession analyses [24]. By including all possible p edic o s in he LASSO, we con olled o con ounding because he LASSO e ained all p edic o s ha we e somehow associa ed wi h he ou come. We did no conside in e ac ions in he models. S a is ical analyses whe e pe o med in S a a e sion 12 (S a aCo p LP, College S a ion, TX, USA [25]) and R Founda ion o S a is ical Compu ing ( e sion 3.3.3). Resul s A i ion among he online pa icipan s in he ini ial phase (ques ion 1a) O e all, 2379 (8.9%) o he in i ed indi iduals consen ed o pa icipa e in HaBIDS (935 pa icipan s in “mixed- mode: pape ”g oup, 750 in “mixed-mode: online”, and 694 in “online-only”). Compa ison o he s udy popula- ion’s composi ion wi h he a ge popula ion’s (inhabi- an s o Lowe Saxony be ween 15 and 69 yea s o age) composi ion showed ha all s a a de ined by age, sex, and educa ion we e occupied in HaBIDS (Addi ional ile 5). Among 2379 pa icipan s, only 109 (4.6%) had no been bo n in Ge many. Da a o 1127 online pa icipan s we e used in he ana- lysis o a i ion in he ini ial phase ( he emaining 317 online pa icipan s did no ill in he i s online ques- ionnai e). One i h o he ini ial online pa icipan s (18.3%, n= 206) discon inued pa icipa ion while 6.3% (n= 71) o mally wi hd ew om he s udy, bo h du ing Rübsamen e al. BMC Medical Resea ch Me hodology (2017) 17:132 Page 5 o 11 he ini ial phase o HaBIDS. Fo 0.9% (n= 10), email de- li e y ailed a some poin du ing he ini ial phase. The emaining 74.5% (n= 840) we e classi ied as egula use s. Figu e 2 shows he cumula i e incidence unc ions. Among all p edic o s, only age showed s a is ically sig- ni ican , uni a iable associa ions wi h bo h ypes o a i- ion (Table 2). The LASSO in compe ing isks eg ession (Table 3) selec ed highe age and less equen In e ne usage as p edic o s o wi hd awal; and younge age, highe PSS, delay in e u ning he consen o m, and ecei ing e- minde emails as p edic o s o discon inua ion. The associa ions be ween age and he ou come had opposi e di ec ions o he wo ypes o a i ion. We ound ha he e was a end ac oss age g oups, wi h 30.0% o he younges pa icipan s (15 o 19 yea s) and 9.9% o he oldes pa icipan s (65 o 69 yea s) lea ing he panel by wi hd awal o discon inua ion (Table 4, p< 0.001 o end). T ans e o pa icipan s o he HaBIDS online panel (ques ions 2a and 2b) O 748 pape -and-pencil pa icipan s and 1319 ini ial online pa icipan s who had been in i ed o he ini ial HaBIDS online panel ( he emaining 312 pa icipan s had o mally wi hd awn so a ), 335 (44.8%) pape -and- pencil pa icipan s and 702 (53.2%) ini ial online pa ici- pan s consen ed o become imely unlimi ed panel membe s. The main easons (in uni a iable analysis) o no pa icipa ing in he ex ended phase we e “no ime” (12.9% o all non-pa icipan s; mul iple choice was possible), “no access o he In e ne ”(8.2%), “ oo many ques ionnai es pe yea ”(6.4%), and “no in e es ”(5.0%). Fo pape -and-pencil pa icipan s, he LASSO in logis- ic eg ession selec ed lowe PSS sco e, highe ID- Sc een, and ha ing e u ned bo h pape -and-pencil ques ionnai es du ing he ini ial phase ins ead o only one as p edic o s o ag eeing o con inue wi h online pa icipa ion (Table 5). Fo ini ial online pa icipan s, he LASSO selec ed inc easing age, being unma ied, lowe PSS sco e, highe WHO-5 and ID-Sc een sco es, delay in e u ning he consen o m, numbe o e u ned ques ionnai es, and inc easing o al numbe o eminde emails du ing he ini ial phase as p edic o s o con inued pa icipa ion (Table 5). A i ion in he ex ended phase (ques ions 1b and 2c) Mo e han one hi d (37.3%, n= 125) o he o me pape -and-pencil pa icipan s did no ill in he i s on- line ques ionnai e a e hei ag eemen o con inue wi h he ex ended phase (i.e. ques ionnai e L) compa ed o 11.0% o he all- ime online pa icipan s (n= 77, p< 0.001 o he compa ison). Only h ee pa icipan s wi hd ew du ing he ex ended phase; hese pa icipan s we e excluded om he ollowing analyses. Among all pa icipan s who illed in ques ionnai e L (n= 835), a - i ion in he ex ended phase was nea ly equal be ween o me pape -and-pencil pa icipan s and all- ime online pa icipan s (5.7% s. 5.8%, espec i ely, p= 0.91) (Fig. 3). A i ion was highe among he “online-only”pa ici- pan s a e he i s i e online ques ionnai es o he ini- ial phase (AB o E) compa ed o a i ion among he o me pape -and-pencil pa icipan s in he ex ended phase (12.4% s. 5.7%, espec i ely, p= 0.004). Fig. 2 Cumula i e incidence cu es o discon inua ion and wi hd awal among he online pa icipan s in he ini ial phase Rübsamen e al. BMC Medical Resea ch Me hodology (2017) 17:132 Page 6 o 11 The LASSO in Cox eg ession selec ed younge age and ecei ing eminde emails as p edic o s o discon- inua ion (Table 6). Discussion We in es iga ed a i ion in a longi udinal panel abou heal h and ound simila p edic o s o a i ion in di e - en phases o he su ey; howe e , he p edic o s di e ed by ype o a i ion (wi hd awal s. discon inua ion). Need o sending a eminde was he s onges p edic o o discon inua ion, bu no p edic o o wi hd awal. Sociodemog aphic cha ac e is ics only played a mino ole o bo h ypes o a i ion. To ou knowledge, he HaBIDS panel is he i s s udy o in es iga e a i ion in di e en phases o a panel. The need o sending a eminde was he s onges p edic o o discon inua ion in he ini ial as well as in he ex ended phase. I s e ec size was e en la ge in he ex- ended (HR 4.20) han in he ini ial phase (HR 1.44). This di e ence be ween phases could be caused by changes in gene al ac o s like commi men , habi s, and panel a igue (desc ibed by Lug ig [8]). Lug ig claims ha commi men is especially impo an in he s a ing phase o a longi udinal s udy when some pa icipan s a e no eally con inced o he s udy. I pa icipa ion i - sel , i.e. illing in he i s ques ionnai es, does no change hei commi men quickly, hese pa icipan s a e likely o discon inue ea ly. Gill e al. [6] also obse ed he impo ance o commi men in hei online mon hly dep ession esc eening p og am o simila leng h as ou s udy and wi h sho ques ionnai es e e y mon h. Ou analyses showed ha he need o ecei ing a eminde o he i s ques ionnai e p edic s a i ion a la e ques- ionnai es (Gill e al. did no send eminde emails, so Table 2 Uni a iable haza d a ios o each ype o a i ion in he ini ial phase Discon inua ion Wi hd awal HR (95% CI) p- alue HR (95% CI) p- alue Age a baseline (pe 10 yea s inc ease) 0.81 (0.73, 0.89) <0.001 1.24 (1.04, 1.47) 0.02 Sex Male Re e ence Re e ence Female 0.91 (0.69, 1.21) 0.54 0.84 (0.52, 1.34) 0.46 Ma i al s a us Ma ied Re e ence Re e ence Unma ied 1.32 (0.97, 1.79) 0.08 0.72 (0.41, 1.28) 0.26 Di o ced/widowed 1.18 (0.73, 1.93) 0.50 0.88 (0.38, 2.06) 0.77 Highes comple ed educa ional le el Lowe seconda y educa ion o app en iceship 1.41 (0.52, 3.84) 0.50 2.44 (0.58, 10.31) 0.22 S ill a uppe seconda y school 0.89 (0.62, 1.29) 0.54 1.56 (0.89, 2.76) 0.12 Uni e si y en ance quali ica ion 1.13 (0.81, 1.56) 0.48 1.07 (0.59, 1.95) 0.82 Uni e si y deg ee Re e ence Re e ence F equency o In e ne usage Daily Re e ence Re e ence Less han daily 1.23 (0.91, 1.67) 0.19 2.03 (1.26, 3.27) 0.004 Sel - a ed heal h s a us Fai /poo 0.93 (0.55, 1.59) 0.80 1.57 (0.67, 3.68) 0.30 Good 1.09 (0.81, 1.47) 0.56 1.73 (1.03, 2.92) 0.04 Excellen / e y good Re e ence Re e ence WHO-5 well-being index (pe 10 poin s inc ease) 0.96 (0.89, 1.03) 0.24 0.96 (0.85, 1.09) 0.57 PSS sco e (pe 1 poin inc ease) 1.07 (1.02, 1.12) 0.002 0.97 (0.89, 1.05) 0.46 ID-Sc een (pe 1 poin inc ease) 1.03 (0.99, 1.07) 0.10 1.01 (0.95, 1.08) 0.73 Time be ween in i a ion and e u n o he illed in o med consen o m (pe week) 1.1 (1.04, 1.17) <0.001 0.98 (0.82, 1.16) 0.78 Had o be eminded o ill in he i s online ques ionnai e (A) Yes 1.85 (1.27, 2.7) 0.001 0.82 (0.33, 2.05) 0.68 No Re e ence Re e ence Rübsamen e al. BMC Medical Resea ch Me hodology (2017) 17:132 Page 7 o 11 we un o una ely canno assess his obse a ion in hei s udy). The need o ecei ing a eminde migh be in lu- enced by commi men , bu also by ime cons ain s o he pa icipan . A e se e al ques ionnai es, pa icipa ion may ha e become a habi , and pa icipan s do no longe con- sciously hink abou he decision o espond, “bu pa - icipa e because hey ha e done so all along”[8]. Lug ig claims ha “once his habi is b oken, he esponden is subsequen ly a a highe isk o […] a i ing”[8]. Con- sen o con inue wi h he online panel in he ex ended phase was associa ed wi h he numbe o p io ques ion- nai es illed in. This could be a p oxy o commi men as well as o habi (based on he pa icipan s esponse be- ha iou , we canno dis inguish hese wo ac o s) and highligh s again he impo ance o hese ac o s in a lon- gi udinal panel. Discon inua ion in he imely unlimi ed HaBIDS on- line panel did no di e by p e ious mode o pa icipa- ion. Howe e , he pe cen age o o me pape -and- pencil pa icipan s who did no e en ill in he i s on- line ques ionnai e in he online panel was highe han among all- ime online pa icipan s. Those pa icipan s who would gene ally ag ee o pa icipa e online, bu a e no used o online da a collec ion o use he In e ne e y i egula ly, p obably d op ou a he i s a emp . Thus, amilia i y wi h he In e ne migh be a mo e im- po an ac o o pa icipa ion in an online panel han a ue mode p e e ence because un amilia i y wi h he In e ne likely leads o a choice o pape -and-pencil mode when his is o e ed. Pa icipan s no using he In e ne daily had a highe isk o wi hd awal, bu no a highe isk o discon inua ion. One explana ion could be ha in e- quen In e ne use s a e mo e selec i e in hei on- line ac i i ies and a he p e e o wi hd aw i hey do no like a s udy. Olde pa icipan s we e mo e likely o wi hd aw han younge ones, bu less likely o become discon inuous use s. O e all, olde pa ici- pan s le he s udy less o en han younge ones. The o me migh be mo e conscien ious so ha hey a he wi hd aw i hey ha e no mo e ime o in e es han discon inue wi hou any eedback o he e- sea che s. Younge pa icipan s migh be busie o ge mo e spam emails han olde ones, which migh esul in discon inua ion ins ead o o mal wi h- d awal. This is also e lec ed by he highe isk o discon inua ion among pa icipan s wi h highe s ess le els (highe PSS sco es). Table 3 Fac o s associa ed wi h isk o a i ion among he online pa icipan s in he ini ial phase Model A: discon inua ion as e en o in e es , wi hd awal as compe ing e en Model B: wi hd awal as e en o in e es , discon inua ion as compe ing e en Be a Haza d a io Be a Haza d a io Age a baseline (pe 10 yea s inc ease) −0.15 0.86 0.10 1.11 F equency o In e ne usage no selec ed in his model Daily Re e ence Re e ence Less han daily 0.28 1.32 PSS sco e (pe 1 poin inc ease) 0.04 1.04 no selec ed in his model Time be ween in i a ion and e u n o he illed in o med consen o m (pe week) 0.05 1.05 no selec ed in his model Had o be eminded o ill in he i s online ques ionnai e (A) no selec ed in his model Yes 0.37 1.44 No Re e ence Va iables we e selec ed using he leas absolu e sh inkage and selec ion ope a o me hod in compe ing isks eg ession PSS: pe cei ed s ess scale (sco e anges om 0 [ e y low s ess le el] o 16 [ e y high s ess le el]) Table 4 Associa ion be ween age and he composi e endpoin (a i ion because o wi hd awal o discon inua ion) in he ini ial phase Age g oup (yea s) Pa icipan s lea ing he panel in he ini ial phase by wi hd awal o discon inua ion 15–19 12 (30.0%) 20–24 21 (28.4%) 25–29 21 (21.0%) 30–34 19 (22.6%) 35–39 27 (28.1%) 40–44 15 (12.6%) 45–49 30 (20.0%) 50–54 21 (15.7%) 55–59 19 (14.6%) 60–64 9 (10.3%) 65–69 10 (9.9%) Rübsamen e al. BMC Medical Resea ch Me hodology (2017) 17:132 Page 8 o 11 Pa icipan s a isk o discon inua ion could be iden i- ied ea ly in he s udy while pa icipan s a isk o wi h- d awal could no . We sugges ha di e en measu es could be applied o he wo g oups. Fo example, an app including ques ionnai es, push eminde s o ill hem in, and elemen s o gami ica ion (applying game mechanics o non-game con ex s in o de o keep pa icipan s en- gaged [26]) migh p e en passi e discon inua ion, bu no ac i e wi hd awal. Too in usi e eminde s could e en s imula e wi hd awal. Depending on he eason o wi hd awal, which we did no assess, eassu ance o da a p o ec ion as well as he s udy’s impo ance o science and socie y and egula eedback wi h s udy esul s migh suppo con inuing pa icipa ion. The same is ue o he discon inue s. Knowing in ad ance ha hose who need mo e suppo in he beginning main ain his a i ude c ea es he oppo uni y o add incen i es. How- e e , hese hypo heses will ha e o be in es iga ed in u- u e s udies. Fu he esea ch abou a i ion in heal h- ela ed panels is needed, o example andomized con- olled ials (RCT) ha examine he e ec o measu es agains discon inua ion among pa icipan s who had o Fig. 3 Kaplan-Meie cu es o discon inua ion (in he ex ended phase) by mode p e e ence Table 5 Fac o s associa ed wi h pa icipa ion in he ex ended phase Model C: pape -and-pencil pa icipan s (asked o swi ch o online mode) Model D: online pa icipan s (asked o s ay in he s udy) Be a Odds a io Be a Odds a io Age a baseline (pe 10 yea s inc ease) no selec ed in his model 0.24 1.27 Ma i al s a us: unma ied s. ma ied no selec ed in his model 0.27 1.31 WHO-5 well-being index (pe 10 poin s inc ease) no selec ed in his model 0.01 1.01 PSS sco e (pe 1 poin inc ease) −0.02 0.98 −0.01 0.99 ID-Sc een (pe 1 poin inc ease) 0.01 1.01 0.04 1.05 Time be ween in i a ion and e u n o he illed in o med consen o m (pe week) no selec ed in his model 0.05 1.05 Numbe o ques ionnai es e u ned du ing ini ial phase a (pe ques ionnai e) 0.19 1.21 0.43 1.54 Numbe o imes a pa icipan had o ecei e eminde emails du ing he ini ial phase (pe eminde ) no applicable o pape -and-pencil pa icipan s −0.27 0.76 Va iables we e selec ed using he leas absolu e sh inkage and selec ion ope a o me hod in logis ic eg ession ID Sc een: mean sco e based on in ec ions and in ec ion-associa ed symp oms in he las 12 mon hs (sco e anges om 0 [no in ec ions in he las 12 mon hs] o 46 [mo e han 42 in ec ions in he las 12 mon hs]) PSS: pe cei ed s ess scale (sco e anges om 0 [ e y low s ess le el] o 16 [ e y high s ess le el]) a Pape -and-pencil pa icipan s ecei ed a o al o wo ques ionnai es; online pa icipan s ecei ed a o al o ele en ques ionnai es Rübsamen e al. BMC Medical Resea ch Me hodology (2017) 17:132 Page 9 o 11