Original Article

Using technology to improve longitudinalstudies: self-reporting with ChronoRecordin bipolar disorder

Bipolar disorder is episodic, recurrent and hetero-geneous (1). Almost all patients have multipleepisodes, with the relapse rate by 5 years estimatedto be 90% (2). The course of bipolar disorder isassociated with both interindividual variation andheterogeneity between patients (3–5). Over a life-time, a patient may experience mixed states, rapidcycling, psychosis, and depressive, manic or hypo-manic episodes. There is great variation betweenpatients in the severity of symptoms, duration ofepisodes, number of episodes, degree of recoveryfrom episodes, pattern of polarity of episodes andintervals between episodes. Furthermore, current

Bauer M, Grof P, Gyulai L, Rasgon N, Glenn T, Whybrow PC. Usingtechnology to improve longitudinal studies: self-reporting withChronoRecord in bipolar disorder.Bipolar Disord 2004: 6: 67–74. ª Blackwell Munksgaard, 2004

Objectives: Longitudinal studies are an optimal approach toinvestigating the highly variable course and outcome associated withbipolar disorder, but are expensive and often have missing data. Thisstudy validates patient self-reported mood ratings using a homecomputer-based system (ChronoRecord) with clinician mood ratings onthe Hamilton Depression Rating scale (HAMD) and Young ManiaRating scale (YMRS), and investigates the patient acceptance of thetechnology.

Methods: After brief training, outpatients with bipolar disorder weregiven the software version of an established paper based self-reportingform (ChronoSheet) to install on a home computer. Every day for3 months, patients entered mood, medications, sleep, life events, andmenstrual data. Weight was entered weekly.

Results: Eighty of 96 (83%) patients returned 8662 days of data. Themean days of data returned was 114.7 ± 32.3 SD The mean percentageof days missing for mood data was 6.1% ± 9.3 SD, equivalent tomissing 7.3 day of the 114.7 days. Self-reported ratings were stronglycorrelated with clinician HAMD ratings ()0.683, p < 0.001).

Conclusions: This study demonstrates concurrent validity betweenChronoRecord and HAMD. Patients with bipolar disorder showed highacceptance of a computer-based system for self-reporting of daily data.Automation of data collection can reduce missing data and eliminateerrors associated with data entry. This technology also enables on-goingfeedback for both patient and researcher during a long-term study.

Michael Bauera,b, Paul Grofc, LaszloGyulaid, Natalie Rasgonb,e, TashaGlennf and Peter C Whybrowb

aDepartment of Psychiatry and Psychotherapy,

Charite-University Medicine Berlin, Berlin,

Germany, bNeuropsychiatric Institute and Hospital,

Department of Psychiatry and Biobehavioral

Sciences, University of California Los Angeles

(UCLA), Los Angeles, CA, USA, cDepartment of

Psychiatry, University of Ottawa, Royal Ottawa

Hospital, Ottawa, Canada, dDepartment of

Psychiatry, University of Pennsylvania School of

Medicine, Philadelphia, PA, USA, eDepartment of

Psychiatry, Stanford School of Medicine, Palo Alto,

CA, USA, fChronoRecord Association, Inc.,

Fullerton, CA, USA

Key words: bipolar disorder – ChronoRecord

computer software – longitudinal studies –

methodology – self-reporting – technology

Received 10 March 2003, revised and accepted for

publication 4 September 2003

Corresponding author: Michael Bauer, MD, PhD,

Department of Psychiatry and Psychotherapy,

Charite-University Medicine Berlin, Campus

Charite-Mitte (CCM), Schumannstr. 20/21, 10117

Berlin, Germany.

Fax: +49-30-450-51-79-62;

e-mail: michael.bauer@charite.de

In part presented at the American Psychiatric Association Annual

Meeting, Philadelphia, PA, May 18–23, 2002; NCDEU June 10–13,

2002, Boca Raton, FL; the European Stanley Foundation Meeting,

Freiburg, Germany, September 12–14, 2002; and the 16th Annual

Meeting of the International Group for the Study of Lithium-Treated

Patients (IGSLI), Prague, Czech Republic, September 26–29, 2002.

The ChronoRecord Association is a non-profit organization that

distributes ChronoRecord software at no charge to qualified clinicians

(http://www.chronorecord.org). None of the authors have a financial

interest in the ChronoRecord Association.

Bipolar Disorders 2004: 6: 67–74Copyright ª Blackwell Munksgaard 2004

BIPOLAR DISORDERS

67

research suggests expanding the bipolar continuumto include less overt symptoms such as hyper-thymic traits (6–7). With the wide range ofphenotypes associated with bipolar disorder, com-plex medication regimens are routinely prescribed(8–9). Despite modern treatments, full recovery isinfrequent; incomplete remission and a high degreeof individual variability are expected (10).

With so much variability in bipolar disorder,longitudinal, prospective studies are an optimalapproach to understand the natural course and toevaluate treatment outcomes. Yet longitudinalstudies are often viewed as impractical and expen-sive. Subjects are difficult to find or to motivate toparticipate for the long-term, and results aredelayed until the end of the study. Several paper-based self-reporting instruments have been used fordata collection in longitudinal studies of patientswith bipolar disorder including the Life ChartMethodology (11), STEP-BP Mood Chart (12),and the ChronoSheet (13). One problem withpaper forms is that patients often complete themafter the fact, just before a visit, and retrospectiverecall of events may be inaccurate and biased (14).Capturing information from paper documents isexpensive, labor intensive and slow. Overall dataquality is negatively impacted by data entry errors.

Frequently, longitudinal data suffers from signi-ficant missing values, differential attrition, meas-urements at unequal time intervals and imbalancednumbers of observations for each participant (15).

Recent acceptance of computer technology bythe general public may offer a mean values toreduce missing data and improve data quality inlongitudinal studies. ChronoRecord was developedto automate the ChronoSheet paper form for self-reporting of both manic and depressive symptomswhich was previously validated against two obser-ver-rated scales (13), the Young Mania Ratingscale (YMRS) (16) and Hamilton DepressionRating scale (HAMD) (17).

The purpose of this study was to validate the self-ratings of mood entered into ChronoRecord soft-ware on a home computer by patients with bipolardisorder with clinician ratings of mood taken on thesame day, and to investigate the feasibility andacceptance of computer technology in a longitud-inal study to collect daily self-reported data.

Methods

Ninety-six patients were consecutively recruitedfrom the mood disorders clinics at UCLA, theUniversity of Ottawa and the University of Penn-sylvania. After a complete explanation of thestudy, all patients signed the written consent form

approved by the Institutional Review Board attheir respective institution. All patients met theDSM-IV criteria for bipolar disorder, diagnosed byclinical interview and confirmed with the MINIInternational Neuropsychiatric Interview (18).Other inclusion criteria were: age 18 years or older,daily access to a personal computer and the skillsto use it, and the ability to read and write English.Patients with antisocial personality disorder ordementia were excluded from the study. After oneand a half hour of training, the patients were givenChronoRecord software to install on their homecomputers. For a 3-month period, the patientsentered mood, sleep, menstrual data, psychiatricmedications, and life events daily. Weight wasentered weekly. Patients returned data by e-mail ordiskette.

ChronoRecord

The ChronoRecord data collection software haslarge, colorful icons for mood, medication andsleep data entry functions. ChronoRecord uses a100-unit visual analog scale between the moodextremes of mania and depression on which thepatient marks mood proportionately. During thepatient’s training, anchor points were set by havingthe patient describe the most depressed and mostmanic states they ever experienced. Patients weretold to describe the predominant features of theextreme state in addition to mood, and oftendepicted a mixed episode as the most manicexperience. Instructions to the patient for moodentry were: (a) enter a single rating that bestdescribes their overall mood for the prior 24 h, (b)carefully review the entire 24-h period, (c) try notto let previous day influence how the current day israted, (d) calibrate the rating to the anchor pointsset during enrollment, and (e) try to enter data atthe same time every day. If a patient did not enterdata on one day, it could be entered later.

Statistical analysis

Missing data analysis. The missing data analysiswas performed on both the patients who enrolled inthe study but failed to return any data and on themissing mood data for patients who returned data.

The demographic characteristics of the 16patients who enrolled but failed to return any datawere compared with those of the 80 patients whoreturned data.

For patients who returned data, the pattern ofmissing ChronoRecord mood data was analyzed toassess whether the data was missing at random(MAR) but related to an observed variable other

Bauer et al.

68

than mood, or informationally missing (IM) wheremissing values are related to the patient’s mood.For each patient who returned data, the distribu-tion of the percentage of missing mood days wascalculated. To evaluate if the missing mood datawas MAR, the distribution of missing mood datawas analyzed by demographic characteristics. Todetermine if the missing mood data was IM, mooddata in quartiles was analyzed by the distributionof missing mood data. To support the assumptionthat missing mood data was missing completely atrandom (MCAR), both the MAR and IM hypo-theses must be rejected.

Validation analysis. To assess concurrent validity,the patient self-rating of mood entered in theChronoRecord software and the results of HAMD,and YMRS tests were compared using Pearsoncorrelation, linear regression and general linearmixed model analyses. Comparisons were madefor the same date at four visits over the 3-monthstudy period. The UCLA study site also comparedChronoRecord ratings with paper-based self-ratingson the Beck Depression Inventory (BDI) (19). Allclinician and self-reported mood ratings covered theprior 24 h, rather than a mean value of the prior7 days for HAMD or 2 days for YMRS, to beconsistent with the validation of the paper Chrono-Sheet (13) and to be comparable with the daily self-reported ChronoRecord mood rating. The use of thePearson correlation analysis to measure the concur-rent validity between HAMD, BDI and YMRS withChronoRecord is consistent with other relatedvalidation studies where systematic differences arenot of primary importance (13, 20–21).

Both linear regression and mixed model analysesused HAMD, BDI and YMRS as dependantvariables with ChronoRecord mood as an inde-pendent variable or covariate. The mixed modelanalysis also included individual patients as arandom factor but did not include interaction

effects. SPSS version 10.0 was used to perform allstatistical computations.

At completion of the study, patients were given aquestionnaire to evaluate ChronoRecord, inclu-ding an overall rating on a 5-point scale rangingfrom 1 as not usable to 5 as excellent.

Results

Eighty of the 96 patients returned 8662 days of data.The primary reasons 16 patients did not return anydata were: (a) discontinued coming to the clinic, (b)lost access to a computer, (c) changed mind aboutcharting, and (d) moved from the area. The meannumber of days of data returned by the 80 patientswas 114.7 ± 32.3 SD (range 24–189 days). Patientsreturned data by either diskette or e-mail in over 240data transfers without incident.

For mood entries, the mean percentage of daysmissing was 6.1% ± 9.3 SD, equivalent to missing7.3 days of the 114.7 days. For mood entries, of the80 patients, 52 (65%) had <5% of days of datamissing, 67 (83.8%) had <10% of days missing,and 73 (91.3%) <20% of days missing. Patientrates for entry of sleep and medication data weresimilar to that for mood. The mean percentage ofdays missing for sleep entries was 6.7% ± 10.8 SD,equivalent to missing 8.2 days of the 114.7 days,and the mean percentage of days missing formedication entries was 8.9% ± 17.0 SD equivalentto missing 9.4 days of the 114.7 days.

Missing data analysis

The patient demographic and clinical characteris-tics were not significantly different between the 80patients who returned data and the 16 who did not(Table 1).

To determine whether the distribution of miss-ing ChronoRecord mood data was related to ademographic variable, the percentage of missing

Table 1. Characteristics of patients with bipolar disorder using ChronoRecord software who returned data and who returned no data

Patients who returned data (n ¼ 80) Patients who returned no data (n ¼ 16) p-value

Age at start (years) 38.67 ± 10.86 SD 36.44 ± 11.28 SD 0.474*Male : female, n (%) 35 (43.8%) : 45 (56.3%) 10 (62.5%) : 6 (37.5%) 0.170**BP I : BP II 58 (72.5%) : 22 (27.5%) 5 (55.6%) : 4 (44.4%) 0.289**Years of illness 15.4 ± 10.7 SD 15.4 ± 10.4 SD 0.990*Single : married : divorced : other (%) 40 : 39 : 13.8 : 7.5 68.8 : 12.5 : 6.3 : 12.5 0.105**Number of hospitalizations 2.43 ± 3.71 SD 2.19 ± 4.11 SD 0.827*Receiving government disability, n (%) (23) 28.8% 3 (18.8%) 0.411**Years of education 15.2 ± 1.8 SD 15.9 ± 2.0 SD 0.977*Number of medications 3.76 ± 1.82 SD N/ATaking mood stabilizer Y : N 73 (91.3%) : 8 (8.8%) N/A

*Comparison by t-test.**Comparison by Pearson’s chi-square test.

Technology and longitudinal studies

69

mood days was calculated for each patient andcategorized as 0–5%, 6–10%, 11–15%, 16–20%,21–25% and 25% and above. The distribution ofgender, taking a mood stabilizer or not, educationlevel, disabled, diagnosis of bipolar I or II, maritalstatus and ethnicity was compared with the distri-bution of the missing mood days yielding thePearson chi-squared statistics of 6.237 (p ¼ 0.284),4.559 (p ¼ 0.472), 20.065 (p ¼ 0.454), 1.252 (p ¼0.940), 3.898 (p ¼ 0.564), 8.628 (p ¼ 0.125) and10.783 (p ¼ 0.823) respectively. An ANOVA ofpatient age by the distribution of missing mooddays produced a between-group F statistic of 0.121(p ¼ 0.987). Since the distribution of missingChronoRecord mood data is not related to anyof the demographic variables, the missing data arenot MAR.

To determine if the missing ChronoRecord mooddata were related to the patient’s functional level,every ChronoRecord mood rating was categorizedinto quartiles (0–25, 25–50, 51–75 and 76–100) andthen summarized for each patient as a percentagedistribution. An ANOVA was used to analyze thepercentage in each quartile for each patient bythe distribution of missing ChronoRecord mooddata yielding between-group F statistics of 1.410(p ¼ 0.230), 1.682 (p ¼ 0.149), 1.343 (p ¼ 0.256)and.364 (p ¼ 0.872) for first, second, third andfourth quartiles respectively. The percentage ofpatient mood data in each quartile did not varysignificantly with the distribution of missing mooddata. This result indirectly suggests that the missingvalues are not IM. As both the MAR and IMhypotheses are rejected, it is assumed that missingChronoRecord mood data is MCAR.

Validation analysis

Using the data points from the 80 patients whereChronoRecord, HAMD (n ¼ 281), YMRS(n ¼ 283) ratings, and BDI (n ¼ 199), werecollected on the same date, the Pearson correlationcoefficient between HAMD and ChronoRecordwas )0.683 (p < 0.001), between BDI and Chrono-Record was -0.673 (p < 0.001) and between YMRS

and ChronoRecord was 0.395 (p < 0.001).Unequal variances were assumed for all correlationsignificance levels.

Table 2 summarizes the results of estimating alinear regression and Table 3 summarizes theresults of estimating a general linear mixed modelthat includes a patient factor. The results of bothtechniques show significant concurrent validity forHAMD. There was a systematic shift between thelinear regression and mixed model results forChronoRecord and BDI. Fig. 1 plots the actualHAMD and ChronoRecord ratings together withthe results of the linear regression as a solid lineand the results of the mixed model as a dotted line.Fig. 2 presents the same information for BDI.

The correlation, linear regression and the mixedmodel analysis failed to validate ChronoRecordwith YMRS. There were only five YMRS ratingsgreater than 15, and 42 ratings between 10 and15 of 283. In contrast, there were 47 HAMDratings greater than 15 and 105 ratings between 10and 15 of 281. The lack of manic patient days withclinician YMRS ratings suggests that the outpa-tient setting of the study failed to provide enoughdata to demonstrate validity.

Of the 80 patients who returned data, 77 (96%)completed the questionnaire to evaluate Chrono-Record. The mean overall rating from the groupwas 4.65 on the 5-point scale indicating patientsenjoyed the technology.

Discussion

To our knowledge, ChronoRecord is the firstautomated system that uses software on a homecomputer for long-term monitoring of a mooddisorder. There was excellent acceptance bypatients with bipolar disorder of using Chrono-Record software on a home computer to entermood ratings daily. The recent phenomenal growthin the percentage of households with personalcomputers has changed the feasibility of automa-ting longitudinal studies. As of September 2001,56% of the US households had at least onecomputer, and 88% of households with computers

Table 2. Concurrent validity analysis using linear regression

Dependant variable n

Intercept ChronoRecord

Adj. R2Estimated p-value* Estimated p-value*

HAMD 281 25.704 < 0.001 )0.390 < 0.001 0.464BDI 199 45.047 < 0.001 )0.650 < 0.001 0.451YMRS 283 )3.202 < 0.001 0.123 < 0.001 0.153

*Significance level of coefficient using the t-statistic.HAMD ¼ Hamilton Depression Rating scale; BDI ¼ Beck Depression Inventory; YMRS ¼ Young Mania Rating scale.

Bauer et al.

70

subscribe to the Internet (22). Furthermore, two-thirds of the 137 000 individuals surveyed used acomputer at work or school (22). With theMicrosoft Windows� operating system now usedby >90% of the desktops in the US and >85%worldwide (23), software for the public can bedeveloped to run on a standard interface. Over thesame time period, the cost of personal computershas dropped dramatically, with complete systemsnow costing <$500. In short, home computershave become standard household appliances.

One benefit of a study design that has patientsenter data directly into a computer is the elimin-ation of data entry. Data entry is expensive,estimated to constitute 5–10% of the total projectcost (24). Data entry is also slow and error prone.Handwritten documents are the most difficult for adata entry staff to handle, requiring good proce-dures and operator intelligence. The data entrystaff must determine what is useful and what isnot on a partially legible, incomplete or �noisefilled� handwritten document. Additionally, anautomated system will provide built-in error checksat the time data is collected from the patient.

ChronoRecord prevents some data entry errorsand requires that the patient correct others. Forexample, ChronoRecord requires that the patiententer weight once in a week before the mood datais accepted, requires confirmation if the patiententers 0 h of sleep before the sleep entry is acceptedand requires confirmation if the patient enters alarge number of pills before the medication entry isaccepted. ChronoRecord also prevents modifica-tion of previously entered data and data entry for afuture date.

Another benefit of an automated system for datacollection is a reduction in missing data. UsingChronoRecord on a home computer, this studyfound that over 80% of patients had <10% ofmissing daily data. The pattern of missing Chrono-Record mood data was not statistically differentfor patients who self-reported extreme ratings ofeither depression (first quartile) or mania (fourthquartile). These results suggests that patients willuse home computers to enter longitudinal data forclinical or research purposes. In addition, as datacan be collected daily rather than weekly ormonthly, a small number of missing values wouldtypically have a minimal impact on the significancelevel of the calculated statistics. If a largepercentage of the study data is missing, moresophisticated analyses may be required to compen-sate for any introduced bias. Other commonproblems found in longitudinal data, such ascorrelation between repeated measurements forindividual patients and irregularly timed data, canbe resolved using analytical techniques such asgeneral estimating equations and general linearmixed models (25–27).

An automated system can provide on-goingfeedback for both the study participant and the

Table 3. Concurrent validity analysis using general linear mixed modelestimation

Dependantvariable n

Intercept ChronoRecord

Estimated p-value* Estimated p-value*

HAMD 281 20.480 < 0.001 )0.303 < 0.001BDI 199 23.065 < 0.001 )0.451 < 0.001YMRS 283 )5.427 0.011 0.149 < 0.001

*Significance level of coefficient using the t-statistic.HAMD ¼ Hamilton Depression Rating scale; BDI ¼ BeckDepression Inventory; YMRS ¼ Young Mania Rating scale.

5

10

15

20

25

30

35

0 10 20 30 40 50 60 70 80 90 100

ChronoRecord

Linear regression

Mixed Model

Data

HA

MD

Fig. 1. Hamilton Depression Rating Scale (HAMD) versus ChronoRecord mood rating data analysis (n ¼ 281).

Technology and longitudinal studies

71

researcher, allowing the timely use of clinicalinformation captured during a longitudinal study(28). This can help motivate subjects to participatefor the study duration, such as with ChronoRecordmood charts (Fig. 3). The mood charts can be usedas decision support tools for clinicians and printedon-demand as educational materials for patients.The researcher will be able to easily evaluatepartial study results.

An automated system can help to standardizedata collection for studies at multiple centers

enabling the creation of large databases containingvalues that were entered in a consistent and reliablemanner. This study collected data from threecenters and patient self-reported mood ratingsusing the ChronoRecord software were highlycorrelated with clinician ratings of depression usingHAMD, similar to our prior findings of correlationbetween ChronoRecord and HAMD using136 data points from 36 patients (Pearson )0.722,p < 0.001) (29). These findings are consistent withthe methodology and results described in studies

5

10

15

20

25

30

35

40

45

50

0 10 0 30 40 50 60 70 80 90 100

ChronoRecord

BD

I

Linear regression

Mixed model

Data

2

Fig. 2. Beck Depression Inventory (BDI) versus ChronoRecord mood rating data analysis (n ¼ 199).

19Apr 01

25 1 7 13 19 25 31May 01

6 12 18 24 30 6 12Jul 01

18 24 30 5 11 17 23Aug 01

29 4 10 16 22 28 4Oct 01

100

25

50

75

100 134 141 145 147 151 153 144 145 147 147 148 147 150 145 147 149 150 151 152 151 153 153 153 153

19Apr 01

25 1 7 13 19 25 31May 01

6 12 18 24 30 6 12Jul 01

18 24 30 5 11 17 23Aug 01

29 4 10 16 22 28 4Oct 01

100

7

14

19Apr 01

25 1 7 13 19 25 31May 01

6 12 18 24 30 6 12Jul 01

18 24 30 5 11 17 23Aug 01

29 4 10 16 22 28 4Oct 01

10

#6#5#4#3#2#1

#1 Depakote

MedicationLegendNumber

(divalproex) 250 mg#2 Depakote (divalproex)500 mg#3 Celexa (citalopram) 20 mg#4 Clonazepam 1.0 mg#5 Zyprexa (olanzapine) 2.5 mg#6 Zyprexa (olanzapine) 5 mg

Daily mood rating

Weekly weight

Life event indication

Period indication

Hours in bed asleep

Hours in bed awake

Fig. 3. 180 day mood chart from a 34-year-old woman with bipolar I disorder.

Bauer et al.

72

measuring the concurrent validity of paper-basedinstruments for self-reporting depressive mood(20, 30–31).

More data from patients with mania is requiredto evaluate the relationship between YMRS andChronoRecord mood ratings. Within the patientsthat participated in this study, mania as meas-ured by YMRS was a rare occurrence. To showChronoRecord validity with YMRS wouldrequire future studies to either include morepatients from a similar outpatient population orto select patients with mania from an inpatientpopulation.

An additional challenge is the possibility thatpatients might underestimate the severity of amanic episode in a self-reported tool relative toself-ratings for patients in a depressive episode.Although patients with mania have poorer insightinto illness than others with bipolar disorder (32),multiple self-reported paper-based rating scaleshave been validated with clinician ratings ofmania, including severity level (20, 33–34). Resultsfrom validation of the self-reported Internal Statescale (13, 35) and recent investigations into thephenomenology of mania (36–37) have suggestedthat increased activation rather than elevatedmood constitutes the core feature of mania, andthat the scores on both clinician and self-ratinginstruments are highly correlated with activation-related features and not mood (13). Within thisframework, the patient’s anchor point for maniaand subsequent self-ratings of mania reflect acti-vation levels for either euphoric or dysphoricmood. Additionally, patients with an anchorpoint set to reflect a mixed episode will generallyrate mixed episodes in the future since mixedepisodes are not random events and occur moreoften in those who previously suffered mixedepisodes (38).

In conclusion, longitudinal studies are optimalfor disorders with a highly variable course such asbipolar disorder. The widespread acceptance ofcomputer technology can be leveraged to improvethe design of longitudinal studies. Automationof data collection may decrease missing data,improve data quality and expand the number ofsubjects who can participate in a standardizedmanner across centers. In this study, Chrono-Record software was used to collect daily moodratings from patients with bipolar disorder.Patient acceptance was excellent and there wasconcurrent validity between patient self-ratedChronoRecord ratings and clinician HAMD rat-ings. Interim data collected during a longitudinalstudy can be used for clinical decision support andpatient education.

References

1. Muller-Oerlinghausen B, Berghofer A, Bauer M. Bipolardisorder. Lancet 2002; 359: 241–247.

2. Tohen M, Waternaux CM, Tsuang MT. Outcome inMania. A 4-year prospective follow-up of 75 patientsutilizing survival analysis. Arch Gen Psychiatry 1990; 47:1106–1111.

3. American Psychiatric Association. Practice Guideline forthe Treatment of Patients with Bipolar Disorder. Am JPsychiatry 2002; 159 (Suppl): 1–50.

4. Judd LL, Akiskal HS, Schettler PJ et al. The comparativeclinical phenotype and long term longitudinal episodecourse of bipolar I and II: a clinical spectrum or distinctdisorders? J Affect Disord 2003; 73: 19–32.

5. Judd LL, Akiskal HS, Schettler PJ et al. A prospectiveinvestigation of the natural history of the long-term weeklysymptomatic status of bipolar II disorder. Arch Gen Psy-chiatry 2003; 60: 261–269.

6. Akiskal HS, Bourgeois ML, Angst J, Post R, Moller H,Hirschfeld R. Re-evaluating the prevalence of and diag-nostic composition within the broad clinical spectrum ofbipolar disorders. J Affect Disord 2000; 59 (Suppl. 1): 5–30.

7. Hirschfeld RM. Bipolar spectrum disorder: improving itsrecognition and diagnosis. J Clin Psychiatry 2001; 62(Suppl. 14): 5–9.

8. Frye MA, Ketter TA, Leverich GS et al. The increasing useof polypharmacotherapy for refractory mood disorders:22 years of study. J Clin Psychiatry 2000; 61: 9–15.

9. Levine J, Chengappa KN, Brar JS et al. Psychotropic drugprescription patterns among patients with bipolar I disor-der. Bipolar Disord 2000; 2: 120–130.

10. Angst J, Sellaro R. Historical perspectives and naturalhistory of bipolar disorder. Biol Psychiatry 2000; 48: 445–457.

11. Leverich GS, Nolen WA, Rush AJ et al. The StanleyFoundation Bipolar Treatment Outcome Network. I.Longitudinal methodology. J Affect Disord 2001; 67: 33–44.

12. Sachs G. Step-BP Blank Mood Chart, Available at http://www.manicdepressive.org/images/samplechart.gif, Acce-ssed on 6/1/2003.

13. Bauer MS, Crits-Christoph P, Ball WA et al. Independentassessment of manic and depressive symptoms by self-rating. Scale characteristics and implications for the studyof mania. Arch Gen Psychiatry 1991; 48: 807–812.

14. Kobak KA, Greist JH, Jefferson JW, Katzelnick DJ,Mundt JC. New technologies to improve clinical trials.J Clin Psychopharmacol 2001; 21: 255–256.

15. Wothke W. Longitudinal and multi-group modeling withmissing data. In: Little TD, Schnabel KU, Baumert Jeds. Modeling Longitudinal and Multiple Group Data:Practical Issues, Applied Approaches and SpecificExamples. Mahwah, NJ: Lawrence Erlbaum Associates,2000: 1–24.

16. Young RC, Biggs JT, Ziegler VE, Meyer DA. A ratingscale for mania: reliability, validity and sensitivity. Br JPsychiatry 1978; 133: 429–435.

17. Hamilton M. A rating scale for depression. J NeurolNeurosurg Psychiatry 1960; 23: 56–62.

18. Sheehan DV, Lecrubier Y, Sheehan KH et al. The Mini-International Neuropsychiatric Interview (MINI): thedevelopment and validation of a structured diagnosticpsychiatric interview for DSM-IV and ICD-10. J ClinPsychiatry 1998; 59 (Suppl. 20): 22–33.

19. Beck AT. Beck Depression Inventory. San Antonio: Har-court Brace & Company, 1996.

Technology and longitudinal studies

73

20. Denicoff KD, Leverich GS, Nolen WA et al. Validation ofthe prospective NIMH-Life-Chart Method (NIMH-LCM-p) for longitudinal assessment of bipolar illness. PsycholMed 2000; 30: 1391–1397.

21. Denicoff KD, Smith-Jackson EE, Disney ER, SuddathRL, Leverich GS, Post RM. Preliminary evidence of thereliability and validity of the prospective life-chart meth-odology (LCM-p). J Psychiatr Res 1997; 31: 593–603.

22. US Department of Commerce, Economics and StatisticsAdministration and National Telecommunications andInformation Administration. A Nation Online: HowAmericans are Expanding their Use of the Internet. Feb-ruary 2002.

23. Woodhead B. Computer Reseller News, 1999; November17. Available at: http://www.itnews.com.au/story.cfm?id¼507, Accessed 10/28/2002.

24. DeMarco T. Controlling Software Projects: Management,Measurements and Estimates. New York: Prentice-Hall,1982.

25. Cnaan A, Laird NM, Slasor P. Using the general linearmixed model to analyse unbalanced repeated measures andlongitudinal data. Stat Med 1997; 30: 2349–2380.

26. Albert PS. Longitudinal data analysis (repeated measures)in clinical trials. Stat Med 1999; 18: 1707–1732.

27. Edwards LJ. Modern statistical techniques for the analysisof longitudinal data in biomedical research. Pediatr Pul-monol 2000; 30: 330–344.

28. Burt T, IsHak WW. Use of technology in outcome meas-urement. In: IsHak WW, Burt T, Sederer LI eds. OutcomeMeasurement in Psychiatry. A Critical Review. Washing-ton, DC: APA, 2002: 383–391.

29. Bauer M, Gyulai L, Glenn T, Whybrow PC. A new clinicaland research tool for bipolar disorder: ChronoRecord

software for daily self-reporting of mood, sleep andmedication. Bipolar Disord 2001; 3 (Suppl. 1): 26.

30. Carroll BJ, Feinberg M, Smouse PE, Rawson SG, GredenJF. The Carroll rating scale for depression. I. Develop-ment, reliability and validation. Br J Psychiatry 1981; 138:194–200.

31. Rush AJ, Gullion CM, Basco MR, Jarrett RB, TrivediMH. The Inventory of Depressive Symptomatology (IDS):psychometric properties. Psychol Med 1996; 26: 477–486.

32. Dell’Osso L, Pini S, Cassano GB et al. Insight into illnessin patients with mania, mixed mania, bipolar depressionand major depression with psychotic features. BipolarDisord 2002; 4: 315–322.

33. Altman EG, Hedeker D, Peterson JL, Davis JM. TheAltman Self-Rating Mania Scale. Biol Psychiatry 1997; 42:948–955.

34. Shugar G, Schertzer S, Toner BB, Di Gasbarro I. Devel-opment, use, and factor analysis of a self-report inventoryfor mania. Compr Psychiatry 1992; 33: 325–331.

35. Bauer MS, Vojta C, Kinosian B, Altshuler L, Glick H. TheInternal State Scale: replication of its discriminating abil-ities in a multisite, public sector sample. Bipolar Disord2000; 2: 340–346.

36. Cassidy F, Murry E, Forest K, Carroll BJ. Signs andsymptoms of mania in pure and mixed episodes. J AffectDisord 1998; 50: 187–201.

37. Akiskal HS, Hantouche EG, Bourgeois ML et al. Towarda refined phenomenology of mania: combining clinician-assessment and self-report in the French EPIMAN study.J Affect Disord 2001; 67: 89–96.

38. Cassidy F, Ahearn E, Carroll BJ. A prospective study ofinter-episode consistency of manic and mixed subtypes ofbipolar disorder. J Affect Disord 2001; 67: 181–185.

Bauer et al.

74