Institutionen för klinisk neurovetenskap Psykologprogrammet, termin 9-10 Huvudämne: Psykologi Examensarbete i psykologi (2PS026), 30 poäng Vårterminen 2013

Cost-effectiveness of Internet-delivered cognitive behavior therapy for severe health anxiety: A randomized controlled trial using behavioral stress management as control condition Författare: Anders Görling & Markus Ronnheden

Handledare: Erik Hedman, fil.dr., Institutionen för klinisk neurovetenskap Examinator: Professor Bo Melin, Institutionen för klinisk neurovetenskap

1 Institutionen för klinisk neurovetenskap Psykologprogrammet, termin 9-10 Huvudämne: Psykologi Examensarbete i psykologi (2PS026), 30 poäng Vårterminen 2013

Cost-effectiveness of Internet-delivered cognitive behavior therapy for severe health anxiety: A randomized controlled trial using behavioral stress management treatment as control condition Sammanfattning Svår hälsoångest (SHA), även kallat hypokondri, är en vanlig psykisk störning. Obehandlat är det ofta ett kroniskt tillstånd som är kopplat till funktionsnedsättning och höga samhällskostnader. Internetbaserad kognitiv beteendeterapi (ICBT) har visat sig vara lovande som en både effektiv och kostnadseffektiv behandling för SHA, men metoden har ännu inte jämförts med en trovärdig aktiv kontrollbehandling. Syftet med denna studie var att undersöka kostnadseffektiviteten för ICBT för SHA och använde Internetbaserad beteendeorienterad stresshantering (IBSM) som kontrollbehandling. En randomiserad kontrollerad studie utfördes för att utvärdera kostnadseffektiviteten av ICBT (n = 79) jämfört med IBSM (n = 79). Primära utfallsvariabler var hälsoångestsymptom, antalet vunna kvalitetsjusterade levnadsår (QALY) och samhällskostnader för psykisk ohälsa. Ett samhällsperspektiv togs på kostnader och inkrementella kostnadseffektivitetskvoter (ICERs) beräknades med hjälp av bootstrapsampling. ICBT gav upphov till stora förbättringar avseende hälsoångest och var bättre än IBSM. Erhållna ICERs indikerade att ICBT sannolikt är kostnadseffektivt om samhället är villigt att betala 14 418 SEK för ett extra fall av remission, eller om samhället är villigt att betala 51 656 kronor för ett extra QALY. Dessa resultat tyder på att ICBT kan betraktas som en kostnadseffektiv behandling för SHA enligt gängse hälsoekonomiska riktlinjer. Nyckelord: hypokondri, hälsoångest, Internet, kognitiv beteendeterapi, kostnadseffektivitet Abstract Severe health anxiety (SHA), also known as hypochondriasis, is a common psychiatric disorder. Untreated it is often a chronic condition, and associated with functional impairment and high societal costs. Internet-based cognitive behavior therapy (ICBT) for SHA has shown promise as an effective and cost-effective treatment, but it has not yet been compared to a credible active control treatment. The aim of this study was to investigate the costeffectiveness of ICBT for SHA using Internet-based behavioral stress management (IBSM) as control treatment. A randomized controlled trial was conducted evaluating the costeffectiveness of ICBT (n=79) compared to IBSM (n=79). Primary outcome variables were symptoms of SHA, quality adjusted life years (QALYs) gained, and societal costs of psychiatric disability. A societal perspective was taken on costs, and incremental costeffectiveness ratios (ICERs) were calculated using bootstrap sampling. ICBT produced large improvements in SHA symptoms and was superior to IBSM. ICERs suggested that ICBT is likely to be cost-effective if society is willing to pay SEK 14,418 for an additional case of remission, or if society is willing to pay SEK 51,656 for one additional QALY. These results suggest that ICBT for SHA can be considered cost-effective according to prevailing health economic standards. Keywords: cognitive behavior therapy, cost-effectiveness, health anxiety, hypochondriasis, Internet

2

Cost-effectiveness of Internet-delivered cognitive behavior therapy for severe health anxiety: A randomized controlled trial using behavioral stress management treatment as control condition Anders Görling & Markus Ronnheden Introduction For each ailment that doctors cure with medications (as I am told they do occasionally succeed in doing) they produce ten others in healthy individuals by inoculating them with that pathogenic agent a thousand times more virulent than all the microbes-the idea that they are ill. -Marcel Proust (The Guermantes Way) The concept of severe health anxiety and hypochondriasis Hypochondriasis is defined in the Diagnostic and statistical manual, fourth edition (DSM-IV; American Psychiatric Association, 2000) as a preoccupation with fears of having, or the idea that one has, a serious disease, based on the misinterpretations of bodily symptoms. This preoccupation must have persisted for at least six months despite appropriate medical evaluation and reassurance, and cause clinically significant distress or impairment. The DSM-IV states that the prevalence is 1%–5% in the general population and 2%–7% among primary care outpatients. When left untreated it has a relatively low remission rate and is often a chronic condition (Barsky, Fama, Bailey, & Ahern, 1998). The name hypochondriasis, however, is somewhat problematic as it has a misleading etymology and since it perpetuates pejorative connotations of malingering. These critiques are reflected in changes proposed in DSM-5 where hypochondriasis will be replaced by somatic symptom disorder and complemented by illness anxiety disorder, the latter being aimed at patients that have high health anxiety but do not present somatic symptoms (American Psychiatric Association, 2013). In response to the criticisms aimed at the name, the authors of this paper have chosen to use severe health anxiety (SHA) as a synonym for the disorder called hypochondriasis in DSM-IV. This term acknowledges the relationship to other anxiety disorders (detailed below), and it also highlights the dimensional, rather than categorical, nature of the construct (Ferguson, 2009; Hiller, Rief, & Fitcher, 2002; Pilowsky, 1967). SHA has not always been considered an independent disorder. Rather, it was once widely held that these symptoms were secondary to other psychopathologies (Kenyon, 1964). Kenyon concluded that it “does not form a nosological entity, but is rather part of another syndrome, most commonly an affective one”. While this was likely the dominant view at the time, arguments emerged for it to be treated independently (Bianchi, 1971; Pilowsky, 1970), and in 1983 Barsky and Klerman concluded that the relationship to affective disorders could not be established on a basis of empirical research. Salkovskis and Warwick (1986) noted that the behavioral pattern typical of SHA appears to follow a continuum, running from mild concerns about unusual bodily sensations to the preoccupation and fears characteristic of patients diagnosed with hypochondriasis. Rather than SHA being related to affective disorders, they proposed that high levels of anxiety were central in the clinical presentation of these patients, a view that has been increasingly popular since. The relation to other anxiety disorders, especially panic disorder and obsessivecompulsive disorder (OCD), is today emphasized by a number of authors (Abramowitz & Braddock, 2006; Barsky, 1992; Deacon & Abramowitz, 2008; Furer, Walker, Chartier, & Stein, 1997; Noyes, 1999; Salkovskis & Warwick, 1986; Starcevic, Fallon, Uhlenhuth, & Pathak, 1994).

3 Costs of SHA SHA has been associated with functional disability (Barsky, Fama, Bailey, & Ahern, 1998; Fink, Ørnbøl, & Christensen, 2010; Seivewright et al., 2004), occupational disability (Mykletun et al., 2009) and increased medical care utilization (Barsky, Ettner, Horsky, & Bates, 2001; Creed & Barsky, 2004; Noyes et al., 1994). These factors are all likely to contribute in making SHA a costly disorder from a societal perspective. To our knowledge no population based study has evaluated the societal costs of SHA specifically. However, a number of studies have estimated the costs associated with other anxiety disorders (Konnopka, Leichsenring, Leibing, & König, 2009; Smit et al., 2006). Smit and colleagues (2006) investigated the societal costs for five common anxiety disorders, and the average annual per capita excess cost was estimated to 3,587 euros (approximately SEK 31,000). Even though SHA may not be as common as other anxiety disorders, it is still has an estimated prevalence of 1-5% (APA, 2000). Given the close relationship to other anxiety disorders, and the aforementioned association with both disability and increased medical care utilization, it is reasonable to assume that it too entails substantial costs for society. Cognitive behavior therapy treatment of SHA Cognitive behavior therapy (CBT) is a broad term used to describe therapies that incorporate both cognitive and behavioral interventions. There are numerous randomized controlled trials that demonstrate the efficacy of CBT treatments in a number of psychiatric conditions, especially anxiety disorders (Butler, Chapman, Forman, & Beck, 2006; Öst, 2008). Many applications of CBT in the treatment of SHA share a common foundation in the aforementioned conceptualization by Salkovskis and Warwick (1986), which stated that there are three loosely associated reasons for patients with SHA to seek medical consultation: (a) the handicap, inconvenience and physical discomfort arising directly from the symptoms; (b) anxiety and intrusive thoughts about the possible cause of the problem, especially catastrophic interpretations of the nature of the symptoms; and (c) discomfort at the possible negative consequences of not taking further action, such as seeking consultation. They argued that the latter two give rise to a range of avoidance behaviors (particularly medical consultation and reassurance seeking) which resembles obsessional behavior. From that assumption they suggested that the prevention of reassurance seeking, coupled with a proper explanation of the role of avoidance behaviors, should be an effective treatment for such patients. Warwick and Salkovskis (1990) developed their theory and presented a maintaining model of SHA with three factors: (a) Cognition: intrusive thoughts about the possible causes of symptoms and the long-term adverse consequences of these causes; (b) Behavior: repeated reassurance-seeking, in combination with efforts to avoid anxiety-provoking cues, maintains the preoccupation, anxiety and bodily sensations; and (c) Physiology: patients with SHA likely perceive both more and augmented physiological symptoms, and these can trigger illness related thoughts. Using this model they proposed two major components of CBT treatment: (a) identification and modification of automatic thoughts and dysfunctional assumptions about health through the use of techniques derived from cognitive therapy for anxiety (c.f. Beck, Emery, & Greenberg, 1985); and (b) Modification of abnormal behavior that can serve to maintain SHA, especially persistent requests for reassurance when it functions similarly to rituals in obsessive-compulsive disorder. In more recent years Warwick and Salkovskis have emphasized the importance of both cognitive (misinterpretations of bodily sensations and illness beliefs) and behavioral (safety-seeking behaviors) aspects as targets for therapeutic intervention (Warwick & Salkovskis, 2001). In a factor analytic study of SHA symptoms, Hiller and colleagues (2002) found that the best subscales for discriminating among hypochondriacs, non-hypochondriacal somatizers

4 and psychiatric controls, were those measuring disease phobia. Based on this finding, Furer and Walker (2005) presented a treatment approach for SHA which focused particularly on the disease phobia aspect. They advocate a view presented by Barlow, Allen and Choate (2004), which suggests that three components are particularly important in maximizing the effectiveness when treating emotional disorders: (a) altering cognitive appraisals, (b) preventing emotional avoidance, and (c) facilitating action tendencies not associated with the emotion that is dysregulated. Furor and Walker argued that one of the most powerful ways to achieve these changes in SHA is through exposure, and present this as the core component of their treatment. The efficacy of CBT in treating SHA has been demonstrated in several studies, many of which were randomized controlled trials (RCTs). CBT has shown to be superior to passive control conditions in a number of studies (e.g. Clark et al., 1998; Seivewright et al., 2008; Sørensen, Birket-Smith, Wattar, Buemann, & Salkovskis, 2011; Warwick, Clark, Cobb, & Salkovskis, 1996). CBT has also showed superiority to psychodynamic psychotherapy in one study (Sørensen et al., 2011), and yielded treatment response comparable to pharmacotherapy (paroxetine) in another (Greeven et al., 2009). One study (Clark et al., 1998) compared CBT to behavioral stress management (BSM, detailed below), and reported that the two treatments produced comparable results. In addition to conventional CBT protocols, recent studies have also showed success in treating SHA with mindfulness-based cognitive therapy (MBCT; Lovas & Barsky, 2010; McManus, Surawy, Muse, Vazquez-Montes, & Williams, 2012). As far as we know (cf. Thomson & Page, 2007), only two RCTs have compared CBT for SHA with another psychological treatment. Sørensen et al. (2011) compared CBT to shortterm psychodynamic psychotherapy (STPP), and to waiting-list. They found CBT to be superior to both, but found no difference between STPP and waiting-list. Clark et al. (1998) conducted a study with the purpose of evaluating CBT for SHA in comparison with an equally credible, but less specific, alternative treatment. They reported considerable difficulties in designing a control treatment that patients found acceptable, but after a considerable amount of pilot work they decided to use BSM. Both treatments were superior to waiting-list, and comparisons between the treatments showed that CBT was more effective than BSM in reducing health anxiety symptoms at mid- and post-treatment. One year later, while symptoms remained significantly lower compared to waiting-list, the two therapies no longer differ from each other. Behavioral stress management for SHA Stress management is a broad term used for various interventions aimed at reducing stress. Interventions in such disparate areas as problem-solving, self-management skills, relaxation, interpersonal skills, affective well-being, and work performance can all fit under the label of stress management (cf. Jones & Johnston, 2000), illustrating its breadth. Cognitive-behavioral stress management can be used as a term for interventions that attempt to reduce stress through the use of cognitive and behavioral techniques such as psychoeducation, relaxation, cognitive restructuring, problem-solving as well as other forms of behavioral coping skills (Brown & Vanable, 2008). In this paper BSM is used to describe the subset of cognitive-behavioral stress management interventions that focuses predominantly on behavioral, rather than cognitive, interventions in order to reduce stress. To our knowledge, only one study have investigated the efficacy of BSM in the treatment of SHA. The aforementioned study by Clark and his colleagues (1998) presented BSM to their participants with the rationale that some people react to stress by becoming worried about their health, and that such worries are best dealt with by acquiring a comprehensive set of stress management skills. Their BSM protocol included psychoeducation about stress, applied relaxation (Öst, 1987), exposure, problem solving,

5 assertiveness training, time-management, and “worry time” (cf. Borkovec, Wilkinson, Folensbee, & Lerman, 1983). It was shown to be effective in reducing symptoms of health anxiety. The contents of the protocol used by Clark et al. (1998) is illustrative of how stress management interventions often use techniques common to CBT treatments, and as such BSM can indeed be considered a form of CBT as well. However, in BSM these techniques aim to reduce stress in general, rather than targeting the thoughts and behaviors hypothesized to be of importance in the CBT-conceptualizations detailed above. Internet-based psychological treatments In the past decade a new form of CBT, Internet-based CBT (ICBT), has emerged. As illustrated by Andersson (2009) there is not one single definition of ICBT. In this paper ICBT is used as a term for therapist-guided treatments that are based on the same theories and conceptualizations as conventional CBT, but employ modern information technology (i.e. the Internet) instead of face-to-face contact as the primary method of delivery. ICBT has been shown to be effective in treating anxiety disorders (Cuijpers, van Straten, & Andersson, 2008; Spek et al., 2007), including SHA (Hedman et al., 2012), and comparisons to conventional CBT has shown that ICBT produce equivalent effects for a number of psychiatric and functional disorders (Hedman, Ljótsson, & Lindefors, 2012). Similar to the case of ICBT, there is no established definition of Internet-based BSM (IBSM). This paper uses IBSM as a term for therapist-guided BSM treatments that employ modern information technology as the primary method of delivery. To our knowledge no study has investigated the effectiveness of IBSM in reducing symptoms of SHA. There are, however, studies suggesting that stress management protocols can be effective when delivered over the Internet (Ljótsson et al., 2011; Richards, Klein, & Austin, 2006; Zetterqvist, Maanmies, Ström, & Andersson, 2003), and that IBSM can be a credible control-treatment to patients (Ljótsson et al., 2011). In addition to the growing body of evidence suggesting the efficacy of Internet-based psychological treatments, these treatments are also a promising way of increasing the general availability of evidence-based psychological treatments. Differences in availability between geographical regions (e.g. Shapiro, Cavanagh, & Lomas, 2003) can be largely remedied through the use of the Internet, and the amount of therapist time needed in ICBT is often less than 20% of that used in conventional CBT (Hedman et al., 2011), reducing costs and making large scale implementations feasible. Cost-effectiveness As societal resources available for health-care are limited, the use of treatments that yield desirable outcomes at low costs is in the interest of both providers and patients. One way of estimating the costs of implementing a new treatment, and to put the costs in relation to its potential benefits and harms, is to use cost-effectiveness analysis. In a cost-effectiveness analysis the incremental costs and effects of one treatment are compared to those of an alternative – usually another treatment or a waiting-list control. The evaluation is usually done from one of two perspectives: one can either take the health-care perspective, where only the direct costs of treatment implementation are considered; or one can take the societal perspective, where all costs associated with the illness and treatment implementation are considered, regardless of when and where they occur (Statens beredning för medicinsk utvärdering [SBU], 2013). The societal perspective is often considered the preferred way of evaluating cost-effectiveness as no special interests are favored (Läkemedelsförmånsnämndens allmänna råd, 2003; SBU, 2013).

6 When conducting a cost-effectiveness analysis it is common to adopt either a disorderspecific outcome such as clinically significant improvement (cf. Jacobson & Truax, 1991), or a generic outcome such as one additonal quality adjusted life-year (QALY; Drummond, Sculpher, & Torrance, 2005). The use of a generic outcome measure, such as QALY, is a special case of cost-effectiveness analysis usually referred to as cost–utility analysis. In this paper we use the term cost-effectiveness analysis to refer only to analyses based on disorderspecific outcomes, while the general term cost-effectiveness can pertain to cost–utility analysis as well. A QALY of 1 is equivalent to one year with full health (i.e. maximum quality of life), and a score of 0 is equivalent with death. In short, a treatment that yields an increase in QALY by .04 over a month is considered to be of equal utility as a treatment that yields an increase in QALY by .01 over four months (cf. Bravo Vergel & Sculpher, 2008, for an accessible introduction). The results of the cost-effectiveness analysis and the cost–utility analysis are often presented as incremental cost-effectiveness ratios (ICERs; Weinstein & Stason, 1977), which is the ratio between difference in cost and difference in effect (SBU, 2013). Depending on the outcome measure used, ICERs can be interpreted as, for example, the cost of achieving one additional case of clinically significant improvement, or the cost of one additional year in full health. As mentioned above, the cost-effectiveness of a treatment is always evaluated in relation to an alternative, and thus a treatment is never considered cost-effective in itself. Rather, it is more or less cost-effective than the treatment it is being compared to. A treatment that produces additional effects at a lower cost than its alternative is always considered more cost-effective. However, when increased effects are seen in tandem with increased costs, the willingness to pay for those additional effects dictate whether the treatment is considered more cost-effective or not. This can be understood as a threshold value – if a treatment yields additional effects at a cost lower than what society is willing to pay, it is considered costeffective. Even though QALY is a universal unit that allows for comparisons between treatments, the willingness to pay for one additional QALY is ultimately an arbitrary decision. Some have suggested $50,000 for one additional QALY as a threshold for cost-effectiveness (Hirth, Chernew, Miller, Fendrick, & Weissert, 2000), whereas others (e.g. Laupacis, Feeny, Detsky, & Tugwell, 1992) have suggested grading the cost-effectiveness of treatments as either strong, moderate or weak. While willingness to pay varies between societies and their healthcare systems, some institutions have explicit guidelines in relation to cost–utility. As an example, The National Institute of Clinical Excellence (NICE) presumes that interventions with an ICER of less than £20,000 per QALY gained are cost-effective (National Institute of Clinical Excellence, 2008). In Sweden, The National Board of Health and Welfare (Socialstyrelsen) considers SEK 100,000 (and below) per QALY a low cost and SEK 500,000 (and above) per QALY a high cost (SBU, 2013). When using a disorder-specific outcome, such as symptom reduction, estimating willingness to pay becomes even more difficult as it does not allow for comparisons across disorders. There are, for example, no standard unit for measuring anxiety in different anxiety disorders (cf. Konnopka, et al., 2009). The main advantage of using a disorder-specific of outcome measure is that it is more sensitive to therapeutic intervention, and sometimes it can also be easier to interpret. Cost-effectiveness of psychological treatments for SHA We have found only two studies investigating the cost-effectiveness of psychological treatments for SHA, one on CBT and one on ICBT. Seivewright et al. (2008) noted reduced medical costs for patients receiving CBT. However, the cost-reduction was not statistically

7 significant, nor did the authors present a thorough evaluation of the cost-effectiveness of their treatment protocol. When it comes to ICBT, one study (the only one, as far as we know) has been published about the effects of ICBT for SHA (Hedman et al., 2012), and this study also evaluated the cost-effectiveness of the treatment. Results indicated that ICBT can be an effective treatment for SHA, and a significantly less costly intervention than traditional CBT. The cost-effectiveness of this treatment, which used a societal perspective, was highly favorable when compared to no treatment. However, the major shortcoming of the study was that it did not include an appropriate control condition. One can therefore not preclude that the observed improvements were the result of non-specific factors involved in regular contact with a therapist, rather than the specific properties of this method of psychotherapy. The lack of adequate control conditions has been identified as a common shortcoming of studies evaluating treatments of SHA (Thomson & Page, 2007). While the efficacy of conventional CBT in treating SHA has been demonstrated in a number of studies, evidence for the efficacy and cost-effectiveness of ICBT is still scarce. The results from the study by Hedman and colleagues (2012) are promising, both in terms of treatment efficacy and cost-effectiveness, but their RCT lacked a credible control group. Since Clark and his colleagues (1998) showed that BSM was credible to patients we argue that Internet-delivered BSM would have been a suitable control condition. In summary, SHA is a common condition that causes substantial suffering for affected individuals and entails high societal costs. Exposure-based ICBT has showed promise as an effective and cost-effective treatment, and its low delivery cost makes it particularly interesting from a cost-effectiveness perspective. To date, ICBT has only been evaluated in comparison with a waiting-list control. In order to better understand its efficacy and costeffectiveness it is important to also investigate ICBT in comparison to an equally credible, but less specific, control treatment. Based on prior evidence IBSM is likely to be a both credible and effective treatment that lacks the specific components hypothesized to be effective in ICBT. Aim of the study The aim of the present study was to prospectively evaluate the cost-effectiveness of ICBT in relation to IBSM, using a societal perspective. We hypothesized that ICBT would be the most cost-effective of the two.

Method Design The present study was a cost-effectiveness study comparing the cost-effectiveness and cost–utility of ICBT with IBSM from a societal perspective. It used prospectively collected data from a RCT. In the trial, participants were randomized to ICBT (n=79) or IBSM (n=79) in a 1:1 ratio with no restrictions or matching. Both treatments were 12 weeks long and participants were assessed at pre-and post-treatment. The trial was pre-registered at clinincaltrials.org (ID: NCT01673035). Recruitment and participants Participants were recruited by self-referral. The opportunity to participate was announced in nationwide newspapers and information about the study was sent to health care units in Stockholm. The participants had to meet the following criteria in order to be eligible for participation:

8 a) fulfill the DSM criteria for hypochondriasis as assessed using the Anxiety Disorder Interview Schedule (ADIS; Di Nardo, O’Brien, Barlow, Waddell, & Blanchard, 1983) b) agree to not undergo any other psychological treatment for the duration of the study c) have no history of CBT in the past four years d) have no serious somatic disease as assessed in the diagnostic interview e) have maintained constant dosage two months prior to treatment if on prescribed medication for anxiety or depression, and agree to keep dosage constant throughout the study f) not currently fulfilling the diagnostic criteria for substance abuse as assessed by the MINI g) have no history of psychosis or bipolar disorder h) no severe depression as indicated by either a score >30 on the Montgomery-Åsberg Depression Rating Scale – Self-Report (MADRS–S; Svanborg & Åsberg, 1994) or as a score of 4 or more on the suicide ideation item of MADRS–S i) other comorbid disorders according to the Mini International Neuropsychiatric Interview (MINI) were allowed, but hypochondriasis had to be the primary concern Screening: Potential participants filled out a screening battery on-line, consisting of the following scales: Health Anxiety Interview (HAI; Salkovskis, Rimes, Warwick, & Clark, 2002), Illness Attitude Scale (IAS; Kellner, Abbott, Winslow, Pathak, 1987; Speckens, Spinhoven, Sloekers, Bolk, & van Hemert, 1996), Whiteley Index (Pilowsky, 1967; Speckens et al., 1996), MADRS–S, Alcohol Use Disorders Identification Test (AUDIT; Saunders, Aasland, Babor, & Grant, 1993), Drug Use Disorders Identification Test (DUDIT; Berman, Bergman, Palmstierna, & Schlyter, 2004), Institute of Medical Technology Assessment Questionnaire on Costs Associated with Psychiatric Illness (TIC-P; Hakkaart-van Roijen, Straten, Tiemens, & Donker, 2002) and the EuroQol Questionnaire (EQ-5D; Rabin & de Charro, 2001). Diagnostic assessments: Diagnostic assessments were made by four master-level students majoring in clinical psychology, as well as a licensed psychologist who also acted as supervisor for the students. The potential participants who completed the screening were contacted by telephone for an assessment interview. The interviews were conducted using a protocol consisting of ADIS, Mini International Neuropsychiatric Interview (MINI; Sheehan et al., 1998), and questions pertaining to the criteria mentioned above. Using the information from interviews in combination with screening data, assessors made a recommendation to the supervisor, who then made the final decision whether to include or exclude. Participants: In total 158 participants were included in the study. A description of the characteristics of the sample is presented in Table 1, and the flow of participants through the trial is shown in Figure 1. Interventions Both treatments were designed as self-help texts and were delivered using an Internetbased platform. The Internet platform allowed therapists to control the pace by which new parts of the self-help text were made available, and allowed for communication between participants and their assigned therapist. Both treatments were divided into 12 modules, with homework assignments relating to the content of each module. The treatments were conducted by four master-level students (clinical psychology) and three licensed psychologists. Therapists were instructed to give feedback on homework assignments, to answer any questions the participants might have, and to prompt participants for their homework if more than a week had passed since it had been assigned. As a general

9 Table 1 Demographic data of participants ICBT Gender Women Men Age Mean age (SD) Range Highest achieved educational level Primary ≥ 9 years Secondary ≤ 2 years Secondary = 3 years Tertiary < 3 years Tertiary ≥ 3 years Postgraduate Clinical data SHA duration, years (SD) Age of debut, mean (SD) Comorbid anxiety disorder Comorbid affective disorder

IBSM

Total

64 (81%) 15 (19%)

61 (77%) 18 (22%)

125 (79%) 33 (29%)

41.9 (13.6) 22-75

41.4 (13.2) 21-70

41.7 (13.4) 21-75

1 (1%) 3 (4%) 10 (13%) 10 (13%) 54 (68%) 1 (1%)

3 (4%) 4 (5%) 14 (18%) 13 (17%) 44 (56%) 1 (1%)

4 (3%) 7 (4%) 24 (15%) 23 (15%) 98 (62%) 2 (1%)

13.0 (13.1) 26.9 (11.5) 26 (33%) 11 (14%)

14.1 (13.1) 25.9 (11.4) 27 (34%) 7 (9%)

13.5 (13.1) 26.4 (11.4) 53 (34%) 18 (11%)

ICBT = Internet-delivered cognitive behavioral therapy, IBSM = Internet-delivered behavioral stress management, SD = standard deviation, SHA = Severe health anxiety

rule therapists replied to participants within 48 hours on week days. There was no face-to-face contact and (with rare exceptions) telephone contact between therapists and participants was not used. The ICBT and IBSM protocols were delivered in the same fashion, but the content of the self-help text and the homework assignments differed between the two. The ICBT protocol was based on the treatment model described by Furer & Walker (2005; Furer, Walker, & Stein, 2007). It emphasized avoidance and safety-seeking behaviors as factors maintaining health anxiety, and the main components of the treatment were exposure (introceptive, in-vivo and imaginary), response prevention, and other CBT techniques, used in order to achieve behavioral change. The protocol is explained in further detail in a previous study (Hedman et al., 2012), and has also been shown to be effective when delivered as a face-to-face treatment (Hedman et al., 2010). The contents of modules in the ICBT treatment are described in Table 2. The IBSM protocol emphasized stress as an important moderator of health anxiety symptoms, and instructed participants to practice the use of a set of tools for stress management, including applied relaxation (Öst, 1987). As mentioned in the introduction, BSM treatments sometimes share many components with CBT treatments and can thus be considered a form of CBT themselves. Acknowledging this, we nevertheless refer to this treatment as IBSM throughout this paper for the sake of clarity. We also want to emphasize that the IBSM treatment was designed to have minimal overlap with the ICBT treatment in terms of treatment components, rather than being a direct replication of any previous study (e.g. Clark et al., 1998). An important consequence of this was a lack of in-vivo exposure in the IBSM protocol. The contents of the individual IBSM modules are presented in Table 3. Materials Assessment of symptoms of health anxiety: Symptoms of health anxiety were measured using HAI. HAI is a self-report questionnaire composed by 64 items measuring

10 Applied for participation, n = 467

Completed diagnostic interview, n = 301

Included in the study and completed pre-assessment, n = 158

Not interviewed, n = 166 Withdrew application, n = 55 Did not complete screening, n = 105 Completed screening, n = 6 Excluded, n = 143 Not primary SHA, n = 63 Currently in treatment for SHA, n = 13 Unstable medication, n = 12 Primary depression, n = 12 Previous CBT for SHA, n = 10 Wished to withdraw, n = 10 Bipolar / Schizophrenia, n = 6 Substance abuse, n = 6 Other, n = 11

ICBT, n = 79

IBSM, n = 79

Completed post-assessment, n = 76

Completed post-assessment, n = 74

Figure 1. Flow of participants throughout the study. SHA = Severe Health Anxiety, CBT = Cognitive Behavior Therapy, ICBT = Internet-delivered Cognitive Behavior Therapy, IBSM = Internet-delivered Behavioral Stress Management.

symptoms and attitudes associated with health anxiety. It has been shown to differentiate between health anxiety and other anxiety disorders, including panic disorder, and to be sensitive to change following therapeutic interventions (Salkovskis et al., 2002). Assessment of quality of life: Quality of life was assessed using the EQ-5D. EQ-5D is a self-report questionnaire with good test-retest reliability (Van Agt, Essink-Bot, Krabbe, & Bonsel, 1994) which measures quality of life using five items: mobility, self-care, usual activities, pain/discomfort and anxiety/depression. The five items of EQ-5D was condensed to a continuous index variable using the weights proposed by Dolan (1997). Assessment of costs: Costs associated with psychiatric illness were assessed using TIC-P. TIC-P is a self-report questionnaire that and covers the monthly costs of psychiatric illness in the following domains: direct medical costs (e.g. health care visits and pharmaceutical costs), direct non-medical cost (e.g. time spent in self-help groups), and indirect non-medical cost (costs of unemployment, sick leave, work cutback and domestic productivity loss). Data from TIC-P was combined with price estimates for each domain, specified as follows: The cost of specific health-care services was specified using the official tariffs of the Southeastern Health Care Region of Sweden (Södra Regionvårdsnämnden, 2008); the cost of pharmaceuticals with statutory pricing was specified using FASS.se (www.fass.se), an on-line service that provides the official costs of those pharmaceuticals; and for pharmaceuticals with free pricing the prices of apoteket.se (www.apoteket.se), the state-owned national

11 Table 2 Breakdown of Internet-based cognitive behavior therapy Modules 1-12 1. Treatment Introduction and Psychoeducation I: Education about health anxiety in general. Behaviors of control, prevention and avoidance and their role in maintaining health anxiety. Homework*: Register your behaviors related to health anxiety. Mindfulness exercise I. 2. Psychoeducation II: A biopsychosocial perspective on health. A cognitive behavioral model explaining how health anxiety is maintained. Homework: Fit personal behaviors into the model. Continue registering behaviors and continue practice mindfulness. 3. Maintaining Factors I – Negative thoughts and catastrophic interpretations: Cognitive distortions. Accepting uncertainties. Homework: Register episodes of health anxiety and associated thoughts. Fit these thoughts into the model from module 2. Continue registering behaviors and continue practice mindfulness. 4. Maintaining Factors II – Catastrophic interpretations of bodily sensations: Introduction to exposure therapy. Homework: Interoceptive exposure. Mindfulness exercise II. 5. Maintaining Factors III – Hypervigilance and safety-seeking behaviors: Introduction to response prevention. Homework: Practice response prevention. Continue practicing interoceptive exposure if it still provokes anxiety. Continue practicing mindfulness. 6. Maintaining Factors IV – Avoidance: Introduction to in-vivo exposure. Homework: In-vivo exposure. Continue practicing response prevention and mindfulness. 7. Maintaining Factors V – Thoughts of disease and death: Introduction to imaginary exposure. Homework: Imaginary exposure: Continue practicing response prevention and in-vivo exposure. 8. Improving Treatment Results I: Troubleshooting for in-vivo exposure. Homework: Scheduling exposure exercises. Continue previous exercises where needed. 9. Improving Treatment Results II: Additional examples of troubleshooting. Homework: Continue previous exercises where needed. 10. Improving Treatment Results III: More time for exposure and response prevention. Homework: Continue previous exercises where needed. 11. Maintaining Treatment Gains I: Treatment summary. Homework: Make a summary of your own. 12. Maintaining Treatment Gains II: Continued improvement. Preparing for future setbacks. * Patients are asked to use exercises in real life, corresponding to homework in a face-to-face setting

pharmaceutical retailer of Sweden, were used. The prices of direct non-medical costs were specified using a rough survey of prices among private providers. When estimating costs related to reduced productivity the human capital approach was used (cf. Drummond et al., 2005). This means that monetary losses associated with unemployment, sick leave and work cutback were based on gross earnings for the full duration of the treatment period. Participants’ salaries were estimated based on their educational level. That is, the median income according to Statistics Sweden (Statistiska centralbyrån; www.scb.se) was used for each of the educational strata presented in Table 1 to estimate the salary of each participant. The cost of domestic productivity loss was specified to SEK 85.7 per hour as suggested by Smit and colleagues (2006). The cost of some of the most commonly used health-care services and pharmaceuticals are presented in Table 4. The cost of treatment delivery was equated to the cost of therapists. The time therapists spent communicating with participants was automatically registered by the Internet platform. The number of hours spent on communication was multiplied with the tariff for a visit to a licensed psychologist as specified by the Southeastern Health Care Region of Sweden (Södra Regionvårdsnämnden, 2008). Other measurements: Treatment credibility was measured using a five-item version of the Treatment Credibility Scale (C-scale; Devilly & Borkovec, 2000). The validity of such scales has been previously demonstrated (Borkovec & Nau, 1972).

12 Table 3 Breakdown of Internet-based Behavioral Stress Management Modules 1-12 1. Treatment Introduction and Applied Relaxation I: Education about health anxiety in general. The role of stress in maintaining health anxiety. Instructions and troubleshooting for relaxation exercise 1. Homework*: Practice relaxation. 2. Applied Relaxation II: Instructions and troubleshooting for relaxation exercise 2. Homework: Practice relaxation. 3. Applied Relaxation III: Instructions and troubleshooting for relaxation exercise 3. A biopsychosocial perspective on health. Advice for improving diet and eating habits. Homework: Practice relaxation and change your diet eating habits if needed. 4. Applied Relaxation IV: Instructions and troubleshooting for relaxation exercise 4. Troubleshooting for changing diet and eating habits. Homework: Practice relaxation. 5. Applied Relaxation V: Instructions and troubleshooting for relaxation exercise 5. Homework: Practice relaxation. 6. Stress management I: Strategies for managing stress (what to do when you get stuck, advice for reducing distractions, recovery during your free time). Homework: Apply stress management strategies in your daily life. Practice relaxation. 7. Stress management II: Troubleshooting for stress management strategies. Relaxation exercise using positive mental images – explanation and troubleshooting. Homework: Practice stress management, relaxation and positive mental image exercise. 8. Stress management III: The significance of sleep and physical exercise. Advice for improved sleep. Homework: Apply changes to sleep and physical exercise where needed. Practice stress management, relaxation and positive mental image. 9. The role of attention in health anxiety: Instructions and troubleshooting for attention training exercise. Homework: Practice attention training, lifestyle changes (diet, sleep, physical exercise), positive mental image, stress management and relaxation. 10. Stress management IV: Troubleshooting for lifestyle changes. Further troubleshooting for stress management. Homework: Practice attention training, lifestyle changes, positive mental image, stress management and relaxation. 11. Maintaining treatment gains I: Treatment summary. Homework: Make a summary of your own. 12. Maintaining treatment gains II: Continued improvement. Preparing for future setbacks. * Patients are asked to use exercises in real life, corresponding to homework in a face-to-face setting

Depressive symptoms was measured using MADRS–S, a self-report scale focusing on core depressive symptoms, that has been demonstrated to differentiate between mild, moderate and severe depressive symptomatology (Svanborg & Åsberg, 2001). Finally, the number of messages sent by therapists and participants was automatically registered. This, in combination with time spent, was used as a rough estimate of therapist attention. Procedure Participants were randomized to either ICBT or IBSM in a 1:1 ratio so that each therapist was assigned the same number of participants, equally divided between the two treatment protocols. The randomization procedure was managed by a third party not involved in the study. All measurements were made before and immediately after treatment, except for treatment credibility which was measured at the end of weeks two and eight. The instruments

13 Table 4 Examples of cost estimates Type of cost Health care services General practitioner Company physician Medical specialist Licensed psychologist Psychotherapist Counselor Physiotherapist Home care

Unit

Cost (SEK)

Consultation Consultation Consultation Session Session Consultation Contact Hour

1214 1103 2103 1305 1330 558 344 279

Pharmaceuticals Alvedon (paracetamol) Cipramil (citalopram) Fontex (fluoxetine) Sobril (oxazepam) Imovane (zopiklon) Flunitrazepam

20 pcs, 500 mg 98 pcs, 20 mg 100 pcs, 20 mg 250 pcs, 10 mg 250 pcs, 5 mg 250 pcs, 1 mg

39 767 725 111 174 211

Non-medical services Alternative care Self-help group

Session Hour

370 65

were administered over the Internet, a valid way of administering questionnaires (Ritter, Lorig, Laurent, & Matthews, 2004), using the treatment platform. Statistical analyses Statistical analyses were conducted using SPSS 20.0 (IBM) and STATA 11.1 (Stata Corporation). Linear mixed models were used to investigate the change in health anxiety symptoms over the treatment period. Using HAI as dependent variable the model was specified using time (pre or post), group (ICBT or IBSM) and the interaction time*group as fixed effects, with a random intercept. The same model was used to investigate changes in costs, quality of life, depressive symptoms and treatment credibility, using total costs from TIC-P, the quality of life index, MADRS-S and C-Scale as dependent variables, respectively. Pre to post effect sizes of health anxiety symptoms and depressive symptoms were calculated for the two treatments respectively. They were expressed as Cohen’s d and were calculated using the mean difference divided by the summed standard deviations of the two samples divided by two. The statistical significance of pre to post effect sizes was verified using paired samples t-tests. Pre to post effect sizes of costs and quality of life were not calculated since data were not normally distributed. The effect size of linear mixed model estimates of group*time interactions was calculated using the estimate divided by the product of the standard error of that estimate multiplied by the square-root of 158 (number of participants). Treatment response was defined by the criteria for clinically significant improvement proposed by Jacobson and Truax (1991), using the HAI as outcome measure. They proposed two criteria for clinically significant improvement: (a) the level of functioning subsequent to therapy places the client closer to the mean of the functional population than it does to the mean of the dysfunctional population, and (b) this change should reflect an improvement beyond that which could be expected by measurement error. In order to define mean values for a functional population we used the value of the healthy control group in a study by

14 Salkovskis and colleagues (2002) evaluating the HAI. The mean HAI pre-treatment value of the sample in the present study was used as the value of a dysfunctional population. Chisquare was used to assess whether the number of responders differed between treatments. Missing data were handled using last observation carried forward and all analyses were conducted based on intention-to-treat. The between-group difference in therapist time was tested for statistical significance using the Mann-Whitney U-test. The between-group difference in number of messages sent by therapists and participants was tested for statistical significance using independent samples t-test. The total cost of each treatment was calculated by adding the cost of treatment delivery to the difference between post- and pre-treatment measurements of costs associated with psychiatric illness (i.e. the cost savings), using last observation carried forward. The pre- to post-treatment changes in costs associated with psychiatric illness were tested for statistical significance using Wilcoxon signed ranks test. Cost-effectiveness was evaluated using an ICER. The ICER was calculated using the formula (ΔC1 - ΔC2) / (ΔE1 - ΔE2), where ΔC1 and ΔC2 represent the changes in costs pre- to post-treatment, and ΔE1 and ΔE2 represent the percentage of participants who achieved clinically significant improvement for the two treatments respectively. Furthermore, measurements of total net costs and clinically significant improvement, were also bootstrapped 5000 times, generating a reliable (Efron & Tibshirani, 1993) distribution estimate for the ICER. The medians of bootstrapped variables were used to estimate the cost of one additional case of clinically significant response. Furthermore, a similar analysis was made to evaluate cost–utility: A cost–utility ICER was calculated as detailed above, but the health outcome was changed from clinically significant improvement to quality of life as assessed with the EQ-5D. The aforementioned bootstrapping procedure was also repeated to generate a distribution estimate for the cost– utility ICER, and the medians of bootstrapped variables were used to estimate the cost of one QALY. The bootstrapped values were also used to generate acceptability curves (c.f. Fenwick, Claxton, & Sculpher, 2001; Löthgren & Zethraeus, 2000). Acceptability curves were used to present the probability of ICBT being more cost-effective than IBSM as a function of willingness to pay.

Results Attrition Data loss due to attrition is shown in Figure 1. On average, ICBT participants completed 8.5 (SD=3.5) modules while the mean number of completed modules in the IBSM group was 8.3 (SD=3.3). Summary of treatment effects Linear mixed model estimates of fixed effects are presented in Table 5. Estimates for health anxiety symptoms revealed a significant main effect of time (p < .001), as well as a significant group*time interaction (p = .013) in favor of ICBT. Estimates for quality of life revealed a significant main effect of time (p = .010) but no significant group*time interaction. Estimates for depressive symptoms revealed a significant main effect of time (p < .001), as well as a significant group*time interaction (p = .022) in favor of ICBT. Effect size calculations revealed large pre- to post-treatment effect sizes on health anxiety symptoms as measured by the HAI (d = 1.80) and depressive symptoms as

15 Table 5 Linear mixed model estimates of fixed effects Parameter HAI Intercept Group = ICBT Time = Post ICBT * Post

Estimate (95% CI)

SE

df

t

p

102.95 (97.78 to 108.11) 2.33 (-4.98 to 9.63) -30.24 (-35.44 to -25.04) -9.35 (-16.66 to -2.04)

2.62 3.71 2.63 3.70

240.39 240.39 151.45 150.73

39.26 .63 -11.50 -2.53