Study selection

A total of 11 794 publications were initially retrieved and, after removing duplicates and adding studies from hand searches, 4150 publication citations remained for broad screening. A further 3799 studies were excluded. Pharmacological trials and studies without a randomisation design were the main reasons for exclusion. A total of 95 studies that met the inclusion criteria were thus identified (Fig. 1). ^{Reference Moher, Liberati, Tetzlaff and Altman23} Of these, 72 studies provided data on negative symptom outcome and were included in the meta-analysis.

Cognitive–behavioural therapies (CBT)

A total of 26 studies investigated the effectiveness of CBT in psychosis. ^{Reference Farhall, Freeman, Shawyer and Trauer24–Reference Valmaggia, van der Gaag, Tarrier, Pijnenborg and Slooff53} Of these, two were CBT interventions adapted specifically for treating negative symptoms. ^{Reference Klingberg, Wittorf, Fischer, Jakob-Deters, Buchkremer and Wiedemann28,Reference Klingberg, Wölwer, Engel, Wittorf, Herrlich and Meisner29,Reference Grant, Huh, Perivoliotis, Stolar and Beck39} Highly significant heterogeneity between CBT studies was found, I ² = 73.6%, P<0.001 (online Table DS1). Details of study characteristics are presented in Table DS2. Treatment was offered over a mean of 6 months (range 1.25–18) for a mean of 28 sessions, each session lasting a mean of 66 min (range 45–90).

Overall, evidence indicated CBT to be an effective intervention for negative symptoms (pooled SMD −0.34, 95% CI −0.55 to −0.12, Fig. 2(a)). Compared with TAU, 59% (10/17) of studies reported CBT to be more effective at the end of treatment (pooled SMD −0.43, 95% CI −0.55 to −0.30). Compared with active control, none of 12 studies suggest a benefit of CBT (pooled SDM −0.11, 95% CI −0.26 to 0.04), although 7 reported substantial but equal improvements in both active conditions. For studies that reported long-term follow-up (mean 27 months), CBT was beneficial compared with TAU across 57% (8/14) of the comparisons. None of the studies with active controls reported such effect, except for one using befriending as the active control. ^{Reference Sensky, Turkington, Kingdon, Scott, Scott and Siddle31,Reference Turkington, Sensky, Scott, Barnes, Nur and Siddle54} There were no differences in effects as a result of differences in study quality.

Fig. 2

Forest plots.

(a) Cognitive–behavioural therapy, skills training, occupational therapy and cognitive adaptation therapy; (b) neurocognitive therapies, exercise therapy, art and music therapies, family-based interventions and miscellaneous interventions.

Skills training, occupational therapy and cognitive adaptation training

A total of 17 RCTs using skills training (n = 11), ^{Reference Bartels, Pratt, Mueser, Forester, Wolfe and Cather55–Reference Rus-Calafell, Gutiérrez-Maldonado, Ortega-Bravo, Ribas-Sabaté and Caqueo-Urízar65} occupational therapy (n = 3), ^{Reference Cook, Chambers and Coleman66–Reference Foruzandeh and Parvin68} cognitive adaptation training (n = 2) ^{Reference Hansen, Østergaard, Nordentoft and Hounsgaard69,Reference Quee, Stiekema, Wigman, Schneider, van der Meer and Maples70} or vocational training (n = 1) ^{Reference Bio and Gattaz71} were included. Only one study was designed specifically to treat negative symptoms ^{Reference Rus-Calafell, Gutiérrez-Maldonado, Ortega-Bravo, Ribas-Sabaté and Caqueo-Urízar65} and there was highly significant heterogeneity between studies, I ² = 85.8%, P<0.001 (online Table DS1). Study characteristics are available in Table DS3. Treatment was offered over a mean of 7 months (range 1.5–24) for a mean of 25.5 sessions (range 12–76) with a mean length of 86 min per session (range 45–180). One active control study compared two variations of skills training: social skills and relapse prevention skills. ^{Reference Horan, Rassovsky, Kern, Lee, Wynn and Green58}

Overall 53% (9/17) of the studies favoured the experimental intervention at the end of treatment (pooled SMD −0.44, 95% CI −0.77 to −0.10, Fig. 2(a)). This effect was largely driven by studies using TAU as the control (pooled SMD −0.42, 95% CI −0.56 to −0.28) and was not present in studies using an active control (pooled SMD −0.05, 95% CI −0.30 to 0.19). Stratified analyses indicated a significant effect of skills training (pooled SMD −0.31, 95% CI −0.45 to −0.17) and of occupational therapy (pooled SMD −1.00, 95% CI −1.48 to −0.51) only. In the study that compared different skills-training programmes, no significant differences were reported. ^{Reference Horan, Rassovsky, Kern, Lee, Wynn and Green58} High- and medium-quality studies reported more favourable outcomes (pooled SMD −0.42, 95% CI −0.56 to −0.28) than those of low quality (pooled SMD 0.00, 95% CI −0.26 to 0.26). Of five studies with follow-up negative symptom data, ^{Reference Roberts, Combs, Willoughby, Mintz, Gibson and Rupp57,Reference Dobson, McDougall, Busheikin and Aldous61–Reference Lin, Chan, Shao, Lin, Shiau and Mueser63,Reference Hansen, Østergaard, Nordentoft and Hounsgaard69} three favoured the experimental intervention at 1 month, ^{Reference Lin, Chan, Shao, Lin, Shiau and Mueser63} 3 months ^{Reference Dobson, McDougall, Busheikin and Aldous61} and 6 months. ^{Reference Xiang, Weng, Li, Gao, Chen and Xie62} The benefit of skills training over social milieu therapy that was reported at 3 months was not maintained at 6-month follow-up in one study. ^{Reference Dobson, McDougall, Busheikin and Aldous61}

Neurocognitive therapies

A total of 16 studies used neurocognitive interventions including cognitive remediation (n = 11); ^{Reference Farreny, Aguado, Ochoa, Huerta-Ramos, Marsà and López-Carrilero72–Reference Bellucci, Glaberman and Haslam82} cognitive training (n = 2); ^{Reference Nemoto, Yamazawa, Kobayashi, Fujita, Chino and Fujii83,Reference Twamley, Vella, Burton, Heaton and Jeste84} cognitive rehabilitation (n = 1); ^{Reference Bellucci, Glaberman and Haslam82} neurocognitive therapy (n = 1); ^{Reference Mueller, Schmidt and Roder85} and cognitive enhancement (n = 1). ^{Reference Eack, Mesholam-Gately, Greenwald, Hogarty and Keshavan86} Another study focused on improving cognitive strategies related to attention, verbal memory and planning. ^{Reference Vauth, Corrigan, Clauss, Dietl, Dreher-Rudolph and Stieglitz87} Three interventions studied were designed primarily to improve negative symptom outcomes ^{Reference Gharaeipour and Scott76,Reference Nemoto, Yamazawa, Kobayashi, Fujita, Chino and Fujii83,Reference Vauth, Corrigan, Clauss, Dietl, Dreher-Rudolph and Stieglitz87} (see online Table DS4 for study characteristics). There was highly significant heterogeneity between studies, I ² = 74.2%, P<0.001 (online Table DS1). Across studies, treatment was offered over a mean of 3.75 months (range 2–24) with a mean of 42 sessions (range 16–120), delivered for a mean of 76 min (range 15–150). Seven studies utilised computer technology for treatment delivery ^{Reference Østergaard Christensen, Vesterager, Krarup, Olsen, Melau and Gluud77,Reference Vita, De Peri, Barlati, Cacciani, Deste and Poli78,Reference Holzer, Urben, Pihet and Jaugey80–Reference Bellucci, Glaberman and Haslam82,Reference Eack, Mesholam-Gately, Greenwald, Hogarty and Keshavan86,Reference Vauth, Corrigan, Clauss, Dietl, Dreher-Rudolph and Stieglitz87} including two specifically designed for the amelioration of negative symptoms. ^{Reference Nemoto, Yamazawa, Kobayashi, Fujita, Chino and Fujii83,Reference Vauth, Corrigan, Clauss, Dietl, Dreher-Rudolph and Stieglitz87} Active controls included computer games ^{Reference Ahmed, Hunter, Goodrum, Batten, Birgenheir and Hardison81} and supportive therapy. ^{Reference Eack, Mesholam-Gately, Greenwald, Hogarty and Keshavan86}

Neurocognitive interventions were effective in only a minority of studies (n = 5, 31%) and data show no overall effect (pooled SMD −0.15, 95% CI −0.41 to 0.11, Fig. 2(b)), regardless of control type and/or study quality. No significant differences were found between modes of neurocognitive treatment delivery. ^{Reference Vita, De Peri, Barlati, Cacciani, Deste and Poli78,Reference Vauth, Corrigan, Clauss, Dietl, Dreher-Rudolph and Stieglitz87} Two of five studies with long-term follow-up data ^{Reference Farreny, Aguado, Ochoa, Huerta-Ramos, Marsà and López-Carrilero72–Reference Penadés, Catalán, Salamero, Boget, Puig and Guarch74,Reference Østergaard Christensen, Vesterager, Krarup, Olsen, Melau and Gluud77,Reference Vauth, Corrigan, Clauss, Dietl, Dreher-Rudolph and Stieglitz87} reported a significant effect at 3 months ^{Reference Twamley, Vella, Burton, Heaton and Jeste84} and 9 months, ^{Reference Mueller, Schmidt and Roder85} and another study reported a significant benefit of cognitive remediation at 4- but not at 9-month follow-up, compared with group leisure activities. ^{Reference Farreny, Aguado, Ochoa, Huerta-Ramos, Marsà and López-Carrilero72}

Exercise therapy

Ten RCTs of exercise therapy on negative symptoms ^{Reference Ho, Au Yeung, Lo, Law, Wong and Cheung88–Reference Cassilhas, Cassilhas, Attux, Cordeiro, Gadelha and Telles97} were reported (see online Table DS5 for study characteristics). Exercises investigated were yoga, ^{Reference Ikai, Suzuki, Uchida, Saruta, Tsukinoki and Fujii89,Reference Visceglia and Lewis90,Reference Manjunath, Varambally, Thirthalli, Basavaraddi and Gangadhar92,Reference Varambally, Gangadhar, Thirthalli, Jagannathan, Kumar and Venkatasubramanian94} aerobic, ^{Reference Acil, Dogan and Dogan91,Reference Scheewe, Backx, Takken, Jörg, van Strater and Kroes93} resistance training, ^{Reference Cassilhas, Cassilhas, Attux, Cordeiro, Gadelha and Telles97} structured walking, tai chi ^{Reference Ho, Au Yeung, Lo, Law, Wong and Cheung88} and traditional dance. ^{Reference Kaltsatou, Kouidi, Fountoulakis, Sipka, Theochari and Kandylis96} Of these only one study was designed to measure negative symptoms as a primary outcome. ^{Reference Ho, Au Yeung, Lo, Law, Wong and Cheung88} Significant heterogeneity between studies was found, I ² = 54.8%, P = 0.039 (online Table DS1). Exercises were offered over a mean of 3 months (range 2 weeks to 8 months) with an average of 26.7 sessions (range 8–48) and a mean of 53 min per session (range 40–60).

An effect of exercise on negative symptoms was found (pooled SMD −0.36, 95% CI −0.71 to −0.01, Fig. 2(b)) that was largely the result of four of seven (57%) comparisons with TAU (pooled SMD −0.42, 95% CI −0.76 to −0.09) whereas active control comparisons showed no effect (pooled SMD −0.07, 95% CI −0.37 to 0.23). Overall, lower-quality studies showed greater effects (pooled SMD −0.37, 95% CI −0.74 to −0.00) than higher-quality studies (pooled SMD 0.14, 95% CI −0.42 to 0.14). Resistance training and exercise as active control comparisons were equally effective in treating negative symptoms. ^{Reference Cassilhas, Cassilhas, Attux, Cordeiro, Gadelha and Telles97} Too few studies were available to warrant a comparison of differences in exercise intervention types. None of the three studies with follow-up assessments found any treatment effect at 1 month, ^{Reference Manjunath, Varambally, Thirthalli, Basavaraddi and Gangadhar92} 1.5 months ^{Reference Ho, Au Yeung, Lo, Law, Wong and Cheung88} and 3 months. ^{Reference Varambally, Gangadhar, Thirthalli, Jagannathan, Kumar and Venkatasubramanian94}

Art and music therapies

Seven RCTs were art- or music-based interventions: two fine arts ^{Reference Crawford, Killaspy, Barnes, Barrett, Byford and Clayton98,Reference Montag, Haase, Seidel, Bayerl, Gallinat and Herrmann99} and five music based ^{Reference Talwar, Crawford, Maratos, Nur, McDermott and Procter100–104} (see online Table DS6 for study characteristics) with highly significant heterogeneity between studies, I ² = 94.9%, P<0.001 (online Table DS1). Treatment was offered over a mean of 4.25 months (range 1–12 months), for a mean of 22 sessions (range 10–52) with a mean of 62.5 min (range 45–90) per session.

Considered together, arts-based treatments were not effective at treating negative symptoms with 57% (4/7) demonstrating no effect (pooled SMD −0.14, 95% CI −0.78 to 0.50, Fig. 2(b)). However, whereas fine-arts-based therapies were not advantageous (pooled SMD 0.57, 95% CI 0.41–0.74), sensitivity analysis revealed a distinct benefit of music-based therapies compared with TAU (pooled SMD −0.58, 95% CI −0.82 to −0.33). ^{Reference Gold, Mössler, Grocke, Heldal, Tjemsland and Aarre101,Reference Tang, Yao and Zheng103,Reference Lu, Lo, Sung, Hsieh, Yu and Chang104} There were no differences in end-of-study outcomes as a result of differences in study quality. Three studies with follow-up data reported no significant lasting ^{Reference Montag, Haase, Seidel, Bayerl, Gallinat and Herrmann99,Reference Lu, Lo, Sung, Hsieh, Yu and Chang104} or emerging ^{Reference Crawford, Killaspy, Barnes, Barrett, Byford and Clayton98,Reference Montag, Haase, Seidel, Bayerl, Gallinat and Herrmann99} benefit.

Family-based interventions

Six RCTs investigating family-based interventions reported negative symptom outcomes ^{Reference Bradley, Couchman, Perlesz, Nguyen, Singh and Riess105–Reference Cai, Zhu, Zhang, Wang and Zhang110} (see online Table DS7 for study characteristics) with marginally significant heterogeneity between studies, I ² = 65.4%, P<0.056 (online Table DS1). The mean duration of treatment was 10.9 months (range 2.5–18) with a mean of 23 sessions (range 10–45) and the mean session duration of 97.5 min (range 60–120). ^{Reference Bradley, Couchman, Perlesz, Nguyen, Singh and Riess105,Reference Wang, Zhou, Yu, Qiu and Wang106,Reference Calvo, Moreno, Ruiz-Sancho, Rapado-Castro, Moreno and Sánchez-Gutiérrez108,Reference Cai, Zhu, Zhang, Wang and Zhang110} One study did not provide standardised treatment ^{Reference Wang, Zhou, Yu, Qiu and Wang106} and two did not specify. ^{Reference Buchkremer, Klingberg, Holle, Schulze Mönking and Hornung109,Reference Cai, Zhu, Zhang, Wang and Zhang110} Family interventions were delivered within: multiple family groups, ^{Reference Wang, Zhou, Yu, Qiu and Wang106,Reference Dyck, Short, Hendryx, Norell, Myers and Patterson107,Reference Buchkremer, Klingberg, Holle, Schulze Mönking and Hornung109} single family groups ^{Reference Cai, Zhu, Zhang, Wang and Zhang110} or a combination of both. ^{Reference Bradley, Couchman, Perlesz, Nguyen, Singh and Riess105,Reference Calvo, Moreno, Ruiz-Sancho, Rapado-Castro, Moreno and Sánchez-Gutiérrez108} One study also used individualised sessions of psychotherapy. ^{Reference Buchkremer, Klingberg, Holle, Schulze Mönking and Hornung109} All studies focused primarily on providing psychoeducation; ^{Reference Bradley, Couchman, Perlesz, Nguyen, Singh and Riess105–Reference Buchkremer, Klingberg, Holle, Schulze Mönking and Hornung109} two studies also included large components of social-skills training. ^{Reference Wang, Zhou, Yu, Qiu and Wang106,Reference Cai, Zhu, Zhang, Wang and Zhang110} Only one study was designed to measure negative symptom outcomes. ^{Reference Dyck, Short, Hendryx, Norell, Myers and Patterson107}

No effect of family intervention was detected either individually or overall (pooled SMD −0.19, 95% CI −0.70 to 0.34) regardless of control comparison. There were no differences in end-of-study outcomes as a result of differences in study quality. Neither of the two studies with follow-up reported a significant effect. ^{Reference Buchkremer, Klingberg, Holle, Schulze Mönking and Hornung109,Reference Cai, Zhu, Zhang, Wang and Zhang110}

Miscellaneous interventions

The following unclassified interventions (n = 10) were included: humour therapy (n = 2); ^{Reference Cai, Yu, Rong and Zhong111,Reference Gelkopf, Gonen, Kurs, Melamed and Bleich112} specialised early intervention for FEP (SEI: n = 2); ^{Reference Thorup, Petersen, Jeppesen, Ohlenschlaeger, Christensen and Krarup113,Reference Garety, Craig, Dunn, Fornells-Ambrojo, Colbert and Rahaman114} acceptance and commitment therapy (n = 1); ^{Reference White, Gumley, McTaggart, Rattrie, McConville and Cleare115} body psychotherapy (n = 1); ^{Reference Röhricht and Priebe116} dog-assisted psychological treatment (n = 1); ^{Reference Villalta-Gil, Roca, Gonzalez, Domenec, Ana Escallani and Asensio117} adherence therapy (n = 1); ^{Reference Anderson, Ford, Robson, Cassis, Rodrigues and Gray118} token therapy (n = 1); ^{Reference Gholipour, Abolghasemi, Gholinia and Taheri119,Reference Choi and Medalia120} and motivation approach to learning arithmetic (n = 1). ^{Reference Choi and Medalia120} Body psychotherapy, motivational learning and humour therapy were designed specifically to treat negative symptoms (see online Table DS8 for study characteristics). Heterogeneity between studies was found to be non-significant, I ² = 33.3%, P = 0.174 (online Table DS1). The mean duration of treatment was 2.25 months (range 1–3), excluding both SEI studies that were conducted over a span of 1.5 years ^{Reference Garety, Craig, Dunn, Fornells-Ambrojo, Colbert and Rahaman114} and 2 years. ^{Reference Thorup, Petersen, Jeppesen, Ohlenschlaeger, Christensen and Krarup113} Whereas SEI interventions were intensive over the entire period of 1.5–2 years, for the other studies mean number of sessions offered was 24 (range 8–60) over an average session length of 66 min (range 30–120).

A significant effect of all miscellaneous studies was found (pooled SMD −0.42, 95% CI −0.77 to −0.07, Fig. 2(b)) driven largely by TAU as opposed to active control comparisons (pooled SMD −0.48, 95% CI −0.75 to −0.21; pooled SMD −0.33, 95% CI −0.67 to 0.02, respectively). Higher-quality studies reported greater overall effects (pooled SMD −0.61, 95% CI −0.87 to −0.36) compared with low-quality studies (pooled SMD 0.08, 95% CI −0.37 to 0.52). Compared with supportive counselling, body psychotherapy was found to be more effective and the effect was retained at 4-month follow-up (pooled SMD −0.74, 95% CI −1.35 to −0.13). ^{Reference Röhricht and Priebe116} Token therapy was more effective than TAU (pooled SMD −0.91, 95% CI −1.66 to −0.15) but not active control (exercise). Treatment in an SEI service was more effective than TAU (regular care) (pooled SMD −0.38, 95% CI −0.53 to −0.22) but the effect was not retained after transfer to regular care. ^{Reference Bertelsen, Jeppesen, Petersen, Thorup, Øhlenschlaeger and le Quach121} Compared with TAU, adherence therapy ^{Reference Anderson, Ford, Robson, Cassis, Rodrigues and Gray118} and acceptance and commitment therapy ^{Reference White, Gumley, McTaggart, Rattrie, McConville and Cleare115} were not effective. Token therapy was not more effective on negative symptoms than an exercise active control. ^{Reference Gholipour, Abolghasemi, Gholinia and Taheri119} Overall, medium- and high-quality studies were more likely to report a significant effect at end of treatment (pooled SMD −0.61, 95% CI −0.87 to −0.36) than were low-quality studies (pooled SMD 0.08, 95% CI −0.37 to 0.52).

Additional analyses

The above results raise several other questions for which we conducted the following additional analyses.

Publication bias

Results of the funnel plot used to determine publication bias indicate a large grouping of studies left of the mean, suggesting that those studies reporting negative effects may have been less likely to have been published (online Fig. DS1).