Comparison with earlier studies

Previous systematic reviews and meta-analyses of DUP and outcome of schizophrenia have mainly focused on the short-term outcome. The general conclusion of previous reviews was that short DUP seemed to be related to better outcome, whereas long DUP was related to poorer clinical and functional outcome and quality of life.^{Reference Marshall, Lewis, Lockwood, Drake, Jones and Croudace2,Reference Perkins, Gu, Boteva and Lieberman3} The small correlations between long DUP and poor clinical outcome are similar to earlier meta-analyses.^{Reference Marshall, Lewis, Lockwood, Drake, Jones and Croudace2,Reference Perkins, Gu, Boteva and Lieberman3} Boonstra et al found that over an 8-year follow-up period the correlation between long DUP and negative symptoms increased from r = 0.12 at baseline to r = 0.20,^{Reference Boonstra, Klaassen, Sytema, Marshall, De Haan and Wunderink28} which is in line with our finding that the correlation between DUP and negative symptoms was stronger when follow-up was longer. The correlations between DUP and remission (as defined by Andreasen et al ^{Reference Andreasen, Carpenter, Kane, Lasser, Marder and Weinberger16}) and hospital treatment have not previously been subjected to meta-analysis. The correlation between long DUP and a decline in social functioning seems to be small, but significant also in the long term. We were the first to study DUP and employment in meta-analysis, and found no significant association.

In meta-regression the length of follow-up had a statistically significant effect on the correlation between DUP and negative symptoms, hospital treatment and global outcome. Interestingly, the correlation between long DUP and poor outcome was stronger in longer follow-up periods. This does not prove that long DUP causes poor outcome,^{Reference McGlashan29} but may indicate that participants with longer DUP have a different type of illness or different coping mechanisms for the symptoms, as well as different cognitive performance from some participants with short DUP. For example, it is possible that more severe negative symptoms, as part of an insidious mode of onset, are one reason for longer DUP,^{Reference Compton, Chien, Leiner, Goulding and Weiss30,Reference Compton, Gordon, Weiss and Walker31} and this type of disorder relates to poor prognosis. This association could remain during the course of illness. It is debatable whether shortening the DUP would decrease negative symptoms, as current treatment options do not generally greatly improve the outcome of such symptoms.^{Reference Tandon32}

A previous review suggested that the correlation between DUP and outcome in developing countries was somewhat contradictory as the DUP tended to be longer but remission rates were higher, despite some correlation between long DUP and poor outcome.^{Reference Chiliza, Asmal and Emsley33} Our meta-analysis gives only limited information on differences between countries based on the level of income. There was no study from a low-income country and only a few (from none to four) studies from countries with lower- or higher-middle level of income in each of the outcome categories. However, the correlations between DUP and general symptomatic outcome and positive symptoms seemed to be smaller in countries with a lower income level.

Methodological discussion and risk of bias

Varying degrees of heterogeneity were found between studies in different outcome categories. This possibly reflects the heterogeneity of DUP and the measures of outcome used. The heterogeneity between studies indicates methodological differences and differences in clinical protocols between countries’ treatment units. Both of these could affect not only the estimation of DUP and outcomes, but may also have confounding effects on both of them. The definition of DUP is challenging and the comparison between participants and samples based on the length of DUP is not always exact. The reasons behind the prolonged periods of untreated psychosis may vary. It might be due to poor insight and more severe types of disorder, but equally in some individuals it could reflect good coping abilities and levels of support, allowing people with psychotic symptoms to live without resorting to mental healthcare and without an accurate diagnosis. The characteristics of local health services may also affect the length of DUP.^{Reference Tandon32}

Defining DUP in a standardised fashion could be one way to decrease variability between studies. The use of a proposed definition of DUP, which takes into account not only the duration of symptoms but also their severity, could be a more accurate means of measuring the potential harmful effects of DUP and reducing the levels of heterogeneity found in our study.^{Reference Compton, Gordon, Weiss and Walker31} Unfortunately detailed assessment of DUP in many of the included studies was not possible owing to limited information given on the methods of defining DUP and their reliability. A number of studies did use structured interviews with more than one source of information, which can be considered reliable, and studies that did not clearly define and separate DUP from (for example) duration of untreated illness were excluded.

Different aspects of possible bias need to be considered. Selection bias may have been present in the original studies as participants with better or worse outcomes might have been left out during the several years of follow-up, and perhaps more probably patients with long DUP might have been more difficult to engage during the study. Some of these issues could be minimised in future studies with more synchronised study designs and systematic measurements of exposure and outcome. In observational designs, investigations are at risk of confounding. The DUP most probably affects treatment in a way that might confound the association between DUP and outcomes. Some confounders, such as personal characteristics of temperament, may affect both DUP and treatment adherence.^{Reference Dassa, Boyer, Benoit, Bourcet, Raymondet and Bottai9} Mode of onset is another possible confounding factor.^{Reference Moncrieff34} However, this was taken into account in four of the original samples without significant effect on the correlation between DUP and outcome.^{Reference Penttilä, Miettunen, Koponen, Kyllönen, Veijola and Isohanni7,Reference Ichinose, Nakane, Nakane, Kinoshita, Ohta and Honda35-Reference Chang, Tang, Hui, Lam, Wong and Chan37} Norman et al found no significant correlation between DUP and mode of onset and found significant correlation with outcome and both of them in only two of eight outcome variables,^{Reference Norman, Manchanda, Windell, Harricharan, Northcott and Hassall38} whereas Selten et al found significant association (P<0.001) between mode of onset and poor outcome, but no clear association with DUP and poor outcome (P = 0.08).^{Reference Selten, Veen, Hoek, Laan, Schols and van der Tweel39} However, in forthcoming studies mode of onset or other markers of severity of the early phase of illness should be taken into account.

Owing to the methodological variation between original studies we were unable to analyse systematically the effect of all possible confounders, but using meta-regression variables such as gender and age at onset seemed to have no major confounding effect. The method of meta-regression used is not the most efficient way to study possible confounders, but as original studies rarely studied confounders, this was our only option. Additionally, the possibility of multiple testing and low number of studies in some of the outcome categories should be noted. Therefore, the possible confounding bias should be considered when interpreting the results of this meta-analysis.

Another methodological challenge was the possible correlation between different outcome categories. Owing to our definition of general symptomatic outcome, results from PANSS scores could have been included in both general symptomatic outcome and positive and negative symptoms. This should also be considered when interpreting the results. As the original studies did not study the correlations between different outcomes, it was not possible to analyse the effect of these correlations in our meta-analysis.

Positive study findings are more likely to be reported than negative findings, which may bias the results of meta-analysis of observational studies. The negative finding regarding publication bias needs to be assessed with caution because the method used (Begg’s test) is not very powerful in detecting bias with a small number of included studies. The observed heterogeneity between studies also presents a challenge in synthesising their results. Our approach, using random effects models, can be considered reliable for studying this sort of data. Sample sizes of included studies varied from 23 to 776, with the majority of studies having relatively small samples. Most of the studies had quite high rates of withdrawal during the follow-up period, which increases the risk of selection bias. The possibility of outcome reporting bias was reduced by contacting authors for their unpublished results and data.

Strengths and limitations

The literature search for the meta-analysis was done using several databases. Articles published in languages other than English were included and were translated by professionals. We also used criteria requiring the majority of participants to have been diagnosed with schizophrenia and at least three out of four with schizophrenia spectrum disorder, which is somewhat stricter than the criteria used in previous meta-analyses.^{Reference Perkins, Gu, Boteva and Lieberman3} Some studies with remarkably different definitions of DUP were excluded from the meta-analysis.^{Reference Compton, Gordon, Weiss and Walker31} We were not able to include some of the earliest studies of delayed treatment and long-term outcome in schizophrenia,^{Reference Moncrieff34-Reference Bottlender, Sato, Jager, Wegener, Wittmann and Strauss36} owing to variations in definitions of diagnosis and DUP. Most of the studies published before the 1990s defined DUP as the duration of symptoms prior to treatment and no information was provided as to whether these symptoms were positive or negative and what threshold was defined for their onset.^{Reference Rupp and Fletcher40,Reference Achte41} The end of DUP had also been defined as the beginning of antipsychotic treatment in one sample, in which participants had their first episode of psychosis in the pre-antipsychotics era, leading to DUPs of over 10 years.^{Reference Scully, Coakley, Kinsella and Waddington42} However, the results of these studies were similar to our findings: Rupp & Fletcher found that duration of symptoms (especially if more than 2 years) before treatment decreased the probability of the improvement of outcome,^{Reference Rupp and Fletcher40} and Scully et al concluded that more severe negative symptoms were predicted by longer DUP.^{Reference Scully, Coakley, Kinsella and Waddington42} The exclusion of randomised controlled trials, most notably the OPUS trial, should not cause an underestimation of the correlation between DUP and outcome in this meta-analysis.^{Reference Jeppesen, Petersen, Thorup, Abel, Øhlenschlaeger and Christensen43,Reference Petersen, Thorup, Øqhlenschlaeger, Christensen, Jeppesen and Krarup44}

There are limitations in this study’s design. Possible differences in the original samples may have affected our results and conclusions. There were only 7 to 19 samples in the different outcome categories. The studies also differed largely in terms of the research methods used. As the original studies presented their results with various effect size measures, we were forced to convert them to correlations using previously presented equations.^{Reference Rosenthal, Cooper and Hedges18,Reference Rosenthal, Rosnow and Rubin19} These conversions can create inaccuracy, as the equations for the conversions are based on normal distribution.^{Reference Rosenthal, Cooper and Hedges18,Reference Rosenthal, Rosnow and Rubin19} However, the use of random weights is an advantage and reduces the risk of such inaccuracies. Most of the studies had analysed the association between DUP and symptoms or global outcome. Owing to the small number of studies definitive conclusions should not be made regarding the longitudinal effects of DUP on, for example, employment and quality of life.

Implications

When developing new treatments for schizophrenia it is important to have a broad knowledge of the effects of current treatment options and periods when the illness has not been treated. Possible differences in the lifestyles of patients with long DUP, compared with those with short DUP, could provide new information for use in the treatment of schizophrenia. It has previously been shown that more severe negative and depressive symptoms correlated with decreased quality of life in schizophrenia,^{Reference Tomotake45} and early intervention might not be able to improve the quality of life in this disorder.^{Reference Thorup, Petersen, Jeppesen and Nordentoft46} Therefore, the factors behind our finding that people with long DUP might have as good a quality of life as those with shorter DUP should be studied further. Future studies of DUP and outcome should use more standardised methods for defining both DUP and outcomes. It would also help the generalisation of results in clinical practice if potential confounders were approached more systematically in studies.

The small correlation between long DUP and poor long-term outcome indicates that early intervention and care in psychosis might have some positive effects on the long-term outcome on schizophrenia. However, drawing evidence-based clinical conclusions from analyses using natural experiments and observational designs is problematic as it is neither possible nor ethical to randomise people to different conditions regarding DUP. It is likely that in the long term other factors (such as individual characteristics, social support and treatment factors) are more influential than DUP in determining outcomes. However, the constant, subtle correlation between long DUP and poor outcome dictates that periods of untreated psychosis should be shortened whenever possible.