Severity levels of depression in DSM‐5 and ICD‐10

Three elements are used to define the severity levels of depression in DSM‐5: the number of symptoms, the level of distress caused by the intensity of the symptoms, and the degree of impairment in social and occupational functioning. The severity categorization applies to all depressive disorders, not just major depressive disorder (MDD). Mild depression is specified when “few, if any, symptoms in excess of those required to make the diagnosis are present, the intensity of the symptoms is distressing but manageable, and the symptoms result in minor impairment in social or occupational functioning”. Severe depression is specified when “the number of symptoms is substantially in excess of that required to make the diagnosis, the intensity of the symptoms is seriously distressing and unmanageable, and the symptoms markedly interfere with social and occupational functioning”. The DSM‐5 does not explicitly define moderate depression other than to say that the number of symptoms, intensity of symptoms, and/or functional impairment are between mild and severe.

There are some problems with the DSM‐5 specification of severity levels. The same definition of the severity specifier is used for MDD and persistent depressive disorder. This is a problem, because persistent depressive disorder requires fewer symptoms than does MDD to meet the DSM‐5 diagnostic threshold. Thus, a patient with persistent depressive disorder who experiences the same number of symptoms as a patient with MDD, and with similar levels of functional impairment and distress, may be classified as more severe because the symptom count may be “substantially in excess” of the diagnostic threshold for persistent depressive disorder but not for MDD.

Another problem with the DSM‐5 severity specifier is that the definition of functional impairment is limited to social or occupational functioning. This is inconsistent with the wording of the impairment criterion for the diagnosis of MDD and persistent depressive disorder, which refers to impairment in social, occupational, or other important areas of functioning. Thus, individuals who maintain social contacts, are not expected to be employed, but are unable to function as students or full‐time parents, could be misclassified as less severe than they actually are.

While moderate severity is not specifically defined, the internal logic of the wording of the moderate severity description has a minor flaw. Mild depression requires low levels of symptoms, distress and functional impairment. Conversely, severe depression requires high levels of all three. Thus, moderate depression should be defined as lying between the mild and severe levels in symptoms, distress or functional impairment (not and/or as DSM‐5 defines it).

Finally, two other variables often considered important in discussions about depression severity – suicidality and need for hospitalization – are not considered in DSM‐5’s definition of severity.

What evidence supports the validity of the DSM‐5 approach towards defining severity in this manner? One study from a population‐based registry of twins who experienced a major depressive episode in the year prior to the interview found that the three aspects of the severity specifier – number of symptoms, severity of symptoms, and degree of functional impairment – were significantly, albeit only modestly, correlated47. The authors concluded that the DSM severity construct was multifaceted and heterogeneous.

A study of psychiatric outpatients with a mood disorder48, 84% of whom were in a major depressive episode, found that the number of DSM‐IV symptoms of MDD was weakly correlated with clinicians’ ratings on the Clinical Global Impression (CGI)49 and the Global Assessment of Functioning (GAF)50. Moreover, the severity ratings of some individual symptoms of depression were as highly correlated with CGI and GAF scores as was the total number of depressive symptoms. A small study of psychiatric inpatients with MDD found that the number of MDD criteria was weakly correlated with the Global Assessment Scale51. Kessler et al52 analyzed data from the National Comorbidity Study (NCS) and found that, compared to individuals who reported five or six MDD criteria during their worst episode of depression, individuals who reported seven to nine MDD criteria experienced more psychosocial impairment, more episodes of depression, and greater chronicity. Wakefield and Schmitz53, 54 examined the NCS database as well as another epidemiological survey and suggested that the number of depressive symptoms was less important than the type of depressive symptoms and other features of complicated depression in predicting future occurrence of a major depressive episode, seeking professional help for depression, a history of suicide attempt, and a history of psychiatric hospitalization. Thus, symptom count does not seem to be an adequate indicator of depression severity.

The ICD‐1055 designates three levels of severity – mild, moderate and severe – based on number of symptoms, severity of symptoms, functional impairment, level of distress and, indirectly, type of symptoms. In contrast to DSM‐5, there is no symmetry in the descriptions of the three levels of severity. Mild depression refers to the presence of two or three symptoms that are distressing though the patient is likely to be able to continue with most activities. Moderate depression requires four or more symptoms with the patient having great difficulty to continue with ordinary activities. Severe depression requires “several symptoms that are marked and distressing, typically loss of self‐esteem and ideas of worthlessness or guilt. Suicidal thoughts and acts are common and a number of ‘somatic’ symptoms are usually present”.

As with the definition of the DSM‐5 severity specifier, little research has been done on the ICD‐10 severity specifier, perhaps because the reliability of making the severity distinctions is poor56. Poor reliability is not surprising, due to the impreciseness of the severity level definitions57.

The severity definitions in the official diagnostic systems have not been used in treatment studies. Rather, in almost all those studies, severity is designated by a score on a symptom rating instrument – usually the Hamilton Depression Rating Scale (HAMD)58 or the Montgomery‐Åsberg Depression Rating Scale (MADRS)59. Thus, treatment studies generally do not consider other factors that have been used to characterize severity, such as level of functional impairment, degree of suicidality, or depressive subtype (i.e., presence of melancholic features or psychotic symptoms)60, 61.

Scales measuring the severity of depression

The severity of depression has been most frequently quantified on paper‐and‐pencil and clinician‐administered rating scales. There is variability amongst the instruments in the time frame covered (the two most common time frames being the past one or two weeks), rating guidelines (most scales use Likert‐type ratings based on symptom frequency, persistence or intensity), and item content.

Little research has examined which parameters provide the most valid indicator of depression severity. Is the severity of depression best conceptualized as the number of symptoms (i.e., present or absent), frequency of symptoms (e.g., every day vs. half the days vs. few days), persistence of symptoms (e.g., always present vs. often present vs. sometimes present), or intensity of symptoms (e.g., severe vs. moderate vs. mild)? Williams et al62, in standardizing the scoring of the HAMD, created a grid scoring format to incorporate information regarding symptom frequency/persistence and intensity in the ratings. The only study to examine whether it is important to consider both intensity and frequency constructs found that symptom intensity was a better indicator of severity than symptom frequency63. In developing the Patient‐Reported Outcomes Measurement Information System (PROMIS) depression scale, Pilkonis et al64 reviewed studies comparing alternative response options and concluded that frequency scaling outperformed intensity ratings, though these were not studies of depression ratings. Thus, the most valid rating format of depression severity scales is unsettled, and has been little studied.

Should the content of a severity scale be based on the diagnostic criteria for the disorder, include other symptoms of depression that are not components of the diagnostic criteria (e.g., low motivation), or include symptoms that are frequent in depressed patients but are defining features of other disorders (e.g., anxiety, irritability)? And by what standard should one judge whether one approach or scale is a more valid indicator of severity? Statistical approaches such as item response theory have been used to construct scales65, 66. While instruments derived from this approach may be psychometrically superior to measures based on the diagnostic criteria for MDD, such measures do not include symptoms that have long been considered to be core components of depression, such as appetite and sleep disturbances or suicidality. If a measure of severity is to be utilized for clinical purposes, and not just for administrative outcome measurement, it is important to include vegetative symptoms, as the presence of these symptoms affects medication selection67, and to assess suicidality because of safety concerns.

While there are differences amongst the scales in how they were constructed, their intended purpose, item coverage, and rating guidelines, the one commonality is that the overall severity of depression is represented by the sum of the ratings of the individual items. For all but a few scales, all items on the scale are rated similarly and contribute equally to the total score. A notable exception is the HAMD58, which includes some items rated 0 to 2, and some others rated 0 to 4. To be sure, measures differ in their emphasis on different content domains of depression68. Some measures have been criticized as being multidimensional, because a unidimensional construct of depression severity is better able to demonstrate treatment effects69. However, all scales, even multidimensional measures which yield subscale scores, as well as instruments that were initially intended to screen for depression rather than being used as indicators of severity, derive a total score that has been used to denote the severity of depression.

The score summation approach is based on some assumptions that have not been empirically supported. Adding up item scores to yield a total score as an indicator of overall depression severity assumes that all symptoms are equal indicators of the severity of depression. However, the different symptoms of depression are not similarly correlated with clinicians’ global ratings of severity48. From the psychometric perspective, the rating options of individual items should convey valid information across the entire spectrum of severity70. Thus, severely depressed patients should more frequently receive the highest rating of a symptom than a low or zero rating, whereas mildly depressed patients should more frequently receive ratings indicating mild severity than the highest rating of a symptom. Santor and Coyne70, using item response theory data analytic techniques, demonstrated that some of the items of the HAMD do not meet these assumptions.

In fact, scales based on item frequency ratings are unlikely to meet these assumptions and therefore may not be good measures of severity. For example, the items on the 9‐item Patient Health Questionnaire (PHQ‐9) are rated on a four‐point scale of symptom frequency during the past two weeks: (0=not at all, 1=several days; 2=more than half the days; 3=nearly every day)71. Patients with MDD would be expected to score a 3 for most of the symptoms that are present, because the definition of MDD requires symptom presence for at least two weeks. Because of the ceiling effect, a patient with MDD seen in primary care who continues to work would score similarly to a depressed patient who is hospitalized because of difficulties with self‐care. While there are several studies of the PHQ‐9 using an item response theory approach, these have been of heterogeneous non‐depressed psychiatric, medical or community samples72, 73, 74, 75, 76, 77, 78. We are unaware of any studies evaluating the performance of the PHQ‐9 items in a sample of depressed patients presenting for treatment. We would predict that, in such a sample, some – perhaps many – items of the PHQ‐9 would be highly skewed because of the aforementioned ceiling effect. No studies have examined the impact of different rating guidelines on the operating characteristics of items on a depression scale.

Implicit in the score summation approach is that low level ratings across many symptoms reflect equal severity to high ratings across a fewer number of symptoms. For example, someone who indicates that, in the past week, he/she has infrequently experienced low mood, insomnia, low self‐esteem, guilt, reduced concentration, fatigue, psychomotor slowing, insomnia, reduced appetite, reduced concentration, impaired decision making, and reduced interest in usual activities would be considered at the same level of severity as someone who reports daily depressed mood, guilt, feelings of inferiority, and suicidal thoughts, but denies all somatic and vegetative symptoms of depression. Likewise, when item ratings are based on symptom intensity, a mild intensity rating of many symptoms is considered the same as a severe intensity rating of a more limited number of symptoms.

The score summation approach, in which all items are weighted equally, is not grounded in a specific overriding conceptualization of severity. If illness severity is conceptualized in terms of mortality risk, then one would expect a measure of depression severity to weight more heavily item ratings of suicidal thoughts, hopelessness and psychomotor agitation than ratings of impaired concentration and fatigue. On the other hand, if illness severity is conceptualized in terms of functional impairment, then one might expect items assessing impaired concentration and fatigue to be weighted more heavily than items assessing appetite reduction or guilt. To be sure, some measures assess functional impairment along with symptomatology63, 71, 79, 80, 81. No symptom‐based measure, however, has been constructed by examining the association of individual items with indices of functional impairment and including on the scale only those items that are independently associated with impairment.

Few studies have examined the association between severity ratings of individual symptoms of depression and multiple external indicators of severity. Faravelli et al48 found marked differences among symptoms in their association with CGI and GAF ratings. Moreover, the symptoms with the highest correlations with CGI ratings – such as depressed mood, psychic retardation, impaired concentration, and anhedonia – tended to have the highest correlations with GAF scores.

Most discussions of the problems with depression scales have focused on their limitations as outcome measures69, 82, 83. However, different aspects of outcome measurement may be of interest, and these differences might result in different approaches towards scale construction. Some measures of the severity of depression have been specifically designed to be sensitive to treatment effects59, 84. Some measures are linked to the symptom criteria that are used to diagnose depression71, 79, 85, 86, whereas others assess a broad range of features that patients indicate are most important in measuring outcome80 or assess a range of diagnostic and associated symptoms of depression87. Descriptions of scale construction typically focus on the content of the measure and rarely discuss the reason for choosing the rating format. For example, in developing the Multidimensional Depression Assessment Scale, Cheung and Power68 reviewed the content of fifteen depression scales and how their scale would address a content gap. There was no discussion, however, of rating formats and why a symptom frequency format was chosen for their measure rather than a rating format assessing symptom intensity.

One of the commonly used clinician rated measures of severity, the MADRS, was designed to be particularly sensitive to change in treatment trials59. Items were selected if they were prevalent in the patients at the beginning of treatment (i.e., prevalence greater than 70%), showed the greatest change from baseline to week 4 of treatment, and change in scores from baseline to week 4 on the symptom showed the greatest correlation with change in total scores on the measure. While there is nothing inherently wrong with constructing a measure in this manner for this purpose, this should not be the basis for selecting items on a measure of depression severity, as the resulting scale can be biased towards the inclusion of items that are particularly sensitive to change for the medication(s) studied. The construction of the MADRS was based on response to mianserin, maprotiline, amitriptyline, and clomipramine – medications that are not commonly used today. Using the same approach to construct a measure today, when different medications are prescribed, might produce a scale that only partially overlaps with the items included on the MADRS. In the same vein, the HAMD, which was published more than 50 years ago, has been criticized for including items that are most responsive to the effects of sedating medications such as tricyclic antidepressants88.

So, while there are many rating scales of depression, and several studies examining them, questions remain as to how to judge if one measure is a more valid indicator of depression severity than another measure. Should it be based on psychometric analyses indicating unidimensionality? Would a “better” measure of severity be more highly correlated with indices of impairment? Be more highly correlated with current suicidal ideation? Be more highly predictive of future suicidal behavior? Be more highly predictive of future mortality in general? Be more highly predictive of future course? Be better able to distinguish depressed patients who do and do not require hospitalization? Demonstrate a larger effect size in a treatment study? Have greater discriminative ability between depression and anxiety, and thus be a “purer” measure of depression?

A problem with depression scales: uncertain validity of cutoffs to define severity groupings

Putting aside the question of how to best conceptualize severity and construct a scale, a problem with the existing literature on depression severity is the inconsistency in the cutoff scores on symptom scales used to demarcate levels of severity, particularly severe depression. The use of various cutoff scores to define severity groups makes it difficult to compare the studies on the treatment implications of severity.

DeRubeis et al89 conducted a mega‐analysis of four studies comparing cognitive‐behavioral therapy and medication, and defined severe depression as a cutoff of 20 or more on the 17‐item HAMD. Likewise, the recent mega‐analysis of placebo‐controlled trials of fluoxetine and venlafaxine used a cutoff of 20 to define severe depression90. Both of these studies cited the landmark study by Elkin et al91 to justify their definition of severe depression. However, Elkin et al did not cite empirical evidence for this cutoff and, in fact, did not refer to the patients scoring above 20 on the HAMD in absolute terms (i.e., having severe depression), but instead referred to these patients in relative terms (i.e., having more severe depression than the patients scoring 20 and below).

In Kirsch et al’s92 meta‐analysis of the impact of severity on antidepressant‐placebo differences, the authors noted that the mean baseline HAMD scores of the antidepressant efficacy trials were in the very severe range (i.e., > 23) based on the American Psychiatric Association (APA)’s Handbook of Psychiatric Measures93 for all but two of the 35 studies included in the analysis. In a prior analysis of antidepressant efficacy studies in the Food and Drug Administration (FDA) data base, Khan et al94 divided the studies into three groups based on pre‐treatment HAMD scores (<24, 25‐27, >28) without indicating the basis for using these cutoff scores to define the groups. Fournier et al95 used the thresholds recommended in the APA’s Handbook of Psychiatric Measures93 to define grades of severity on the HAMD (mild to moderate: <18; severe: 19 to 22; very severe: >23). In contrast to these studies, and the APA guidelines, most pharmacotherapy studies have used a cutoff of 25 on the 17‐item HAMD to define severe depression96, 97, 98, 99, 100, 101 and this cutoff has been recommended by several experts102, 103, 104. Thus, severe depression has not been consistently defined.

Fundamental to studies on the treatment implications of severity levels is the validity of the cutoffs on the HAMD to define the severity categories. In none of the discussion sections of the meta‐analyses and pooled analyses of the reports on severity and treatment outcome were questions raised about the cutoffs used to define the grades of severity. The APA’s Handbook of Psychiatric Rating Scales93 cited only two small studies in support of the cutoff scores to identify severity subtypes, and neither study provided support for the APA guidelines. One was a study examining the validity of deriving a HAMD equivalent score on the Schedule for Affective Disorders and Schizophrenia105. This study did not attempt to determine the cutoff scores on the HAMD indicating grades of severity. The second study examined the association between HAMD scores and global ratings of severity in 59 depressed inpatients106. The authors did not derive (or recommend) cutoff scores corresponding to severity levels. Thus, it is unclear why a cutoff of 19 was recommended in the APA Handbook to identify severe depression. The UK National Institute for Health and Clinical Excellence (NICE) guidelines recommended a cutoff of 23 to identify severe depression on the HAMD, though no research was cited to support this recommendation107.

Because of the limited amount of empirical research establishing cutoff scores for bands of severity on the HAMD, and the significance accorded to severity by treatment guidelines, our clinical research group also examined this issue in 627 psychiatric outpatients with MDD who were rated on the CGI108. The cutoff score on the HAMD that maximized the sum of sensitivity and specificity was 17 for the comparison of mild vs. moderate depression and 24 for the comparison of moderate vs. severe depression. Based on a review of the available evidence, as well as the recommendations that a cutoff of 7 be used to define remission, we recommended the following severity ranges for the 17‐item HAMD: no depression (0‐7); mild depression (8‐16); moderate depression (17‐23); and severe depression (>24).

Each of the above studies derived cutoff scores based on clinicians’ global judgments of severity. A limitation of these studies is that it is not known on what basis the global judgments of severity were made. Were some symptoms of depression considered better indicators of severity than other symptoms? For example, are symptoms characteristic of melancholic or endogenous depression given greater weight in clinicians’ CGI ratings? Are clinicians’ global ratings disproportionately influenced by degree of suicidality? Do clinicians consider psychosocial impairment in making their CGI ratings? We are unaware of any studies that have attempted to derive severity ranges on the HAMD, or any other depression scale for that matter, based on degree of impairment or level of suicidality.

Another problem with depression symptom scales: different scales classify patients into different severity groups

In clinical practice, self‐report questionnaires are preferable to clinician‐rated scales because they take less time to administer. If self‐report scales are to be used to classify patients into severity categories, and if treatment recommendations are to be based, in part, on severity classification, then it is important for different scales to classify individuals similarly. However, because the content of measures differ, it would not be surprising if there were significant differences between measures.

Cameron et al109 compared the PHQ‐9 and the Hospital Anxiety and Depression Scale (HADS) severity classifications in a sample of primary care patients referred by their general practitioners in the UK to a mental health worker110. No information was provided regarding the patients’ psychiatric diagnoses. They found that the PHQ‐9 overclassified severity compared to the HADS, with twice as many patients classified in the severe range. Other studies comparing the PHQ‐9 and the HADS in medical patients found similar results111, 112. However, these studies lack an external validator and it is therefore unclear if the PHQ‐9 overclassifies, or the HADS underclassifies, severity. A second study by Cameron et al107 included the second edition of the Beck Depression Inventory (BDI‐II)113 along with the PHQ‐9 and HADS, and also assessed the patients with the HAMD. The participants were primary care patients who had been diagnosed by their general practitioner with depression. Both the PHQ‐9 and BDI‐II overclassified severity compared to the HAMD, whereas the HADS underclassified severity.

We are aware of only one study that compared self‐report scales in a sample of psychiatric outpatients with MDD114. Our clinical research group compared severity classification on three measures that assess the DSM‐IV/DSM‐5 symptom criteria for MDD: the Clinically Useful Depression Outcome Scale (CUDOS)79, the Quick Inventory of Depressive Symptomatology (QIDS)85, and the PHQ‐971. The patients were also rated on the 17‐item HAMD. In a study of depressed outpatients, we found that the correlations between the HAMD and all three self‐report scale scores were nearly identical, and the average correlation among the three self‐report scales was .73. However, the scales significantly differed in their distribution of patients into severity categories. Approximately one‐third of the patients scored in the mild range on the HAMD and CUDOS, whereas approximately 10% of the patients were mildly depressed according to the PHQ‐9 and QIDS. On the CUDOS and HAMD, moderate depression was the most frequent severity category, whereas on the PHQ‐9 and QIDS the majority of the patients were classified as severe. The majority of the patients in the moderate range on the HAMD were in the severe range on the PHQ‐9 and QIDS. Significantly fewer patients were classified as severely depressed on the CUDOS compared to the PHQ‐9 and QIDS.

With the three self‐report measures being highly correlated with each other, and equally correlated with the HAMD, what, then, might account for the marked differences between scales of similar content in the distribution of patients into severity groups?

The cutoffs on the three scales to define the severity groups were derived in different ways, and this was likely responsible for the differences between the scales in severity classification. For example, Kroenke et al71 indicated that the cutoff scores on the PHQ‐9 were chosen for the pragmatic reason of making them easier for clinicians to recall. They also noted that alternative cutoffs did not increase the association between increasing PHQ‐9 severity and indices of construct validity. When selecting the cutoff scores to define the severity ranges on the PHQ‐9, the developers of this questionnaire did not consider the potential impact of the broadness by which severity ranges were defined and how this might impact on treatment recommendations of official treatment guidelines.

Kroenke et al71 indicated that, when severity groupings based on different cutoffs are equally associated with external variables, then the cutoffs can be chosen based on their ease of recall. We disagree with this reasoning. For all scales measuring the severity of depressive symptoms, the thresholds distinguishing patients with mild, moderate and severe depression do not represent well‐demarcated lines separating the severity subtypes. As with other areas of psychopathology, the severity of depression better corresponds to a dimensional than a categorical model of classification115. Thus, alternative cutoffs to categorize severity groupings are likely to also be valid when the groupings are compared on an external variable such as psychosocial functioning. However, one should not be cavalier about the choice of cutoffs, because they impact on the relative broadness of each of the severity categories.

If clinicians are to follow official treatment guidelines’ recommendations and base initial treatment selection on the severity of depression, then it is important to have a consistent method of determining depression severity. The marked disparity between standardized self‐administered scales in the classification of depressed outpatients into severity groups indicates that there is a problem with the use of such instruments to classify depression severity. If official treatment guideline recommendations were followed, then use of measures such as the QIDS and PHQ‐9, which broadly define the severe category, would result in greater reliance on medication in preference to psychotherapy as the first line treatment option for MDD. Caution is thus warranted in the use of these scales to guide treatment selection until the thresholds to define severity ranges have been better established empirically.

The importance of severity of depression in treatment: official guideline recommendations

Notwithstanding the aforementioned problems with conceptualizing the severity of depression, and defining the cutoffs on scales for severity levels, depression severity is an important consideration in treatment decision‐making. The severity of depression has influenced treatment recommendations in official guidelines. The third edition of the APA’s guidelines for the treatment of MDD recommend both psychotherapy and pharmacotherapy as monotherapies for depression of mild and moderate severity, and pharmacotherapy (with or without psychotherapy) for severe depression1. The NICE updated guidelines for the treatment and management of depression discourage the use of antidepressant medication as the initial treatment option for mild depression, and recommend medication together with empirically supported psychotherapy for moderate and severe depression2. As reported by van der Lem et al116, treatment guidelines in the Netherlands also recommend pharmacotherapy as the first treatment option for severely depressed patients, and either pharmacotherapy or psychotherapy for mildly and moderately depressed patients. While the recommendations in these guidelines are not entirely consistent, they are unanimous in recommending medication as the treatment of choice for severe depression.

The treatment significance of severity has been studied in several different ways. There are controlled studies, effectiveness studies, pooled analyses, and meta‐analyses examining the impact of severity on particular treatments117, 118, 119, 120, 121, 122, comparing treatments across a range of severity99, 123, 124, 125, 126, 127, comparing medication and placebo across a range of severity128, 129, comparing psychotherapy and control groups across a range of severity130, 131, comparing treatments amongst severely depressed patients96, 101, 102, 132, and examining whether severity predicts short‐term outcome42, 133, 134, 135, treatment resistance136, longer‐term outcome40, 137, 138, 139, and relapse38.

Severity of depression and pharmacotherapy

In the past decade, questions have been raised whether selective serotonin reuptake inhibitors (SSRIs) and other new generation antidepressants are effective in non‐severe depression. Khan et al94 analyzed 45 clinical trials in the FDA database and found that in studies with a mean baseline 17‐item HAMD score of 24 or less there was little evidence that antidepressant medication was superior to placebo, whereas in studies with a mean baseline HAMD score of 28 or greater there was clear evidence that medication was superior to placebo. Kirsch et al92 similarly examined the FDA database, and they also examined the efficacy of antidepressants as a function of mean baseline HAMD score in the trial. Their results largely replicated the findings of Khan et al94 that drug‐placebo differences were largest in the studies with the highest baseline severity (i.e., HAMD >28). Kirsch et al92 found that antidepressants were significantly more effective than placebo in the less severe cohorts, but they considered the difference in response to be modest and clinically insignificant.

In contrast to the analyses of the FDA database by Kirsch et al92 and Khan et al94, Fournier et al95 pooled individual patient data from six published studies. Kirsch et al and Khan et al used aggregated mean scores for an entire study as the unit of analysis. That is, they compared studies with different mean severity scores at baseline. The problem with this approach is that a group of patients with a mean score in the severe range will also include some patients in the mild and moderate severity ranges. Likewise, a group of patients with a mean score in the mild or moderate severity range will include some patients scoring in the severe range. Pooling individual patient data avoids the problem of severity group misclassification at the individual patient level. Fournier et al95 replicated the finding that drug‐placebo differences were clinically significant only for severely depressed patients, and found only a small effect size for mildly and moderately depressed patients.

More recently, other pooled analyses of patient level data (rather than aggregated data from a trial) have been conducted. Using pharmaceutical company data bases, these analyses included all studies of a product, thereby avoiding the bias inherent in examining only published studies140. The results of three large, pooled analyses of published and unpublished studies, which included between 4,000 and 10,000 subjects each, indicated that antidepressants are effective across a range of severity90, 129, 141. These analyses, and the controversy that has been stirred regarding the efficacy of antidepressants, highlights the impact that considerations of severity might have on clinical practice.

Severity of personality disorders and functioning

Although severity has been defined in various ways in the PD literature, a general consensus appears to have emerged that PD severity is inherently linked with level of maladaptive functioning164, 165, 166, 167, 168, 169. It is widely acknowledged that extreme trait or symptom variation is insufficient to diagnose PDs or to dictate diagnostic severity. Rather, the emphasis lies in having extreme personality traits in the presence of impairment associated with those traits. Unlike physical illnesses, or even depression, which are more focused on symptom presentation, personality diagnoses are intertwined with adaptive functioning. Like depression, PDs by definition must result in “distress or impairment” to be diagnosed33. In contrast to depression, however, the symptom criteria for diagnosing PDs include both affective/cognitive/emotional and functional components. For example, impoverished occupational and financial functioning is included in symptom criteria for antisocial PD, and failure to engage in social and leisure activities is part of the criteria for obsessive‐compulsive PD.

The interrelationship between functional impairment and personality leads many to conclude that PD severity is a combination of extreme personality disturbance and maladaptive functioning associated with that disturbance165, 169. In fact, functioning is so fundamental to determining PD presence and severity that some authors argue that assessing extreme traits/symptoms is unnecessary170, 171, 172, 173. Thus, one need not demonstrate symptom severity if sufficient impairment is judged to be present. However, the dysfunction must be determined as due to the presence of the personality features, even if they are not extreme. For example, using the multiaxial DSM‐IV, Livesley174 proposed defining PD as present diagnostically on Axis I, and coding personality traits separately on Axis II. Widiger and Trull169 proposed a similar model, only using the GAF score on Axis V as a stand in for severity.

Taken together, these models converge on defining severity as a generalized, adaptive failure of an intrapsychic system required to fulfill daily life tasks166. Although specific areas of impairment differ, there is convergence on impairment in three broad areas: identity formation, self‐control (or direction), and interpersonal relationships164. However, some research indicates that pathological personality traits and functioning are so closely intertwined that they may not represent distinct domains175.

Severity of personality disorders as described in DSM‐5 and ICD‐10

There is no clear mention of severity with respect to PDs in the main section II of DSM‐533. However, the overall description of PDs includes severity indicators common to other disorders. For example, PDs are specifically noted to be inflexible, maladaptive, pervasive, and associated with “clinically significant” functional impairment or subjective distress. Functional impairment is an indicator of severity in many physical and psychiatric disorders; pervasiveness is a severity indicator for depression; and subjective distress is identified as indicating a “severe case” for disorders of mood and sexual function. As it stands, there is no official method for indicating PD severity in DSM‐5.

Section III (Emerging Measures and Models) of DSM‐5 includes an alternative model for diagnosing PDs. Diagnosis is defined via a combination of severity levels of dysfunction and elevated personality traits, and severity is determined principally by dysfunction associated with elevated traits33. This model does not designate a measure for overall severity, but “moderate or greater impairment” is required for diagnosis. Impairment is operationalized as falling into one of five levels, with the extreme end indicative of severe personality pathology. The Level of Personality Functioning Scale (LPFS) is proposed to rate impairments in functioning, and therefore also PD severity. Ratings are made for self (identity and self‐direction) and interpersonal (empathy and intimacy) functioning. Levels include: 0 (little or no impairment), 1 (some impairment), 2 (moderate impairment), 3 (severe impairment), 4 (extreme impairment). Individuals with extreme impairment are described as having an impoverished, unclear identity and self‐direction with maladaptive self‐concept, and completely lacking capacity to engage interpersonally.

Interestingly, DSM‐5 Section III also includes discussion of an additional measure of personality traits, the Personality Inventory for DSM‐5176. The items are clearly trait content related; however, the measure provides an overall summed score identified as measuring “overall personality dysfunction”. The identification of extreme traits as indicative of dysfunction is curious, but not inconsistent with the significant overlap between functioning and PD traits/symptoms found elsewhere in the literature175. Nonetheless, this suggests that extreme traits are at least indicative of extreme dysfunction, which is the primary index of severity in this model.

Similar to the DSM‐5, the ICD‐10 does not make mention of severity in PD classifications. However, several papers have been published on changes proposed for ICD‐11, which are substantial. Most notably, the primary classification of PDs will change to one based on severity of personality disturbance. Description of PD traits or features is optional but will not be required for diagnosis3, 4.

Consistent with the larger literature, the proposed changes to the ICD‐11 conceptualize severity primarily as dysfunction, or the personality‐related problems experienced by the individual. Again, five levels of severity are proposed, though they vary slightly from those in the DSM. Summed together, severity levels are dictated first by pervasiveness of the impairment (across situations or limited), and secondarily by the number of problematic personality traits (multiple or single). At the highest level of severity, risk to self or others is also assessed. Thus, the most severe cases are identified by functioning above all else. Symptoms/traits and risk of harm are secondary, but also considered. Unlike the DSM‐5 alternative model proposal, dysfunction in self and identity is not included in severity ratings3, 4. At the time of this writing, the ICD‐11 has not yet been published, and therefore these definitions should be considered provisional. Nonetheless, the emphasis on functioning via severity ratings has been criticized for insufficient research establishing its reliability and validity177.