In this secondary analysis of a large, well-characterized sample of clinically depressed patients, we identify specific objective sleep disturbances that are associated with poor treatment outcome in depression. In particular, we found that objectively measured prolonged sleep latency (>30 minutes) is associated with significantly increased risk of non-remission following pharmacologic and/or psychotherapeutic treatment for depression. Results were independent of baseline clinical characteristics (depression or anxiety symptoms), length of follow-up, treatment modality (psychotherapy alone versus pharmacotherapy with or without psychotherapy) and demographic characteristics (age, sex) which are known to influence treatment outcomes. We also found that increasing numbers of sleep disturbances, particularly those with 3 or more disturbances were 3 times more likely to be non-remitters than those without any sleep disturbances. These findings are consistent with previous evidence linking prolonged sleep latency with adverse physical health outcomes, including risk of developing the metabolic syndrome38
In contrast, subjective insomnia complaints alone were not associated with increased risk of poor treatment outcome. In populations in which insomnia is commonly comorbid, perhaps only the more severe insomnia phenotypes, characterized by objective sleep disturbance, are associated with increased risk for poor outcomes.
Indeed, we found that insomnia in combination with objectively measured prolonged sleep latency predicted increased risk of non-remission. In addition, insomnia and short sleep duration individually and in combination were associated with a significantly increased risk of non-remission. Notably, with regards to both combinations of objective criterion with or without insomnia, the population with the objective marker but lacking the insomnia complaint were the smallest subgroups (n=27 for prolonged sleep latency without insomnia and n=44 for short sleep duration without insomnia). Thus, caution is warranted in interpreting the odds ratios associated with these specific subgroups due to the relatively small sample sizes.
Nevertheless, these findings suggest important avenues for future research as they highlight the heterogeneity in traditional characterizations of sleep disturbances. This is perhaps most striking when considering the consistently reported health risks associated with short sleep duration. Epidemiologic investigations typically define short sleepers on the basis of subjective responses to a single-item question assessing typical sleep duration, without assessing distress associated with short sleep (necessary for identifying insomnia), and without assessing sleep duration objectively. Therefore, the question remains whether the risk associated with short sleep duration is due to the pathophysiological effects of short sleep in the absence of distress (i.e., non-complaining short sleepers), or the combination of short sleep with concomitant distress that may potentiate risk.
Investigation of these sleep subgroups has clear clinical implications as they may be characterized by differential risk trajectories and may require different treatment approaches. For instance, a limited number of studies have sought to characterize “non-complaining short sleepers,” with some evidence suggesting higher rates of subclinical hypomanic symptoms in this subgroup40
. Other studies have shown that non-complaining short sleepers primarily differ from their complaining, short sleeping counterparts, in their lack of psychological distress.41–43
Our exploratory analysis did not identify specific clinical or demographic characteristics that distinguished the subgroup with short sleep duration or prolonged sleep latency, without insomnia (the smallest subgroups). Rather, consistent with our hypothesis that the combination of an objective marker of sleep disturbance with insomnia may represent a more biologically severe phenotype of insomnia, we found that the subjective + objective disturbance groups had significantly higher baseline measures of depressive and anxiety symptoms. These findings are also consistent with Vgontzas and colleagues’ work showing that the combination of insomnia with objectively measured short sleep duration potentiates the risk for adverse outcomes, ranging from neurocognitive functioning to hypertension, diabetes, and mortality.16–18;44
Given robust links between depression and cardiometabolic consequences,45–48
the combination of insomnia with objectively measured sleep disturbances may confer increased risk for poor depression outcomes, as well as accelerated risk trajectories for cardiometabolic morbidity and mortality. Previous evidence also suggests that the combination of insomnia with high stress responsivity may represent a distinct endophenotype of depression49–52
Several inherent limitations associated with this secondary data analysis warrant caution in interpreting the findings. Based on stringent eligibility criteria for each of the included protocols, patients with severe, comorbid psychiatric diagnoses were excluded, which may limit generalizability to the broader population of depressed individuals. Other limits to generalizability concern the fact that the sample was predominantly female and Caucasian, and included a disproportionate number of patients with recurrent depression. In addition, the outcome in the current study, remission status, was based on HDRS scores of ≤7 for two consecutive monthly ratings, as this threshold is one of the most commonly used and recommended criteria for defining remission status in depression treatment studies.35
However, given that the total score includes sleep items, it is possible that non-remission reflects stability of sleep complaints rather than non-remission of depression per se. However, in follow-up analyses that restricted the sample to those with scores on the HDRS sleep items at baseline <4 (29% of the sample), results were unchanged, suggesting that our definition of non-remission was reflecting symptoms other than sleep complaints present at baseline. More generally, pooling data from multiple protocols inherently introduces heterogeneity which may pose a threat to the validity of the findings. However, these threats were mitigated by statistical covariation for patient characteristics that differentiated the individual protocols, by the absence of significant interactions based on treatment modality, and the fact that all patients were selected from the same stable community population, were diagnosed and assessed using standard, reliable measures, and were evaluated and treated at the same institution by affiliated investigators. Regarding the in-laboratory sleep studies, heterogeneity may also have been introduced due to differences in the protocols for the timing of sleep recordings (based on fixed lab time or according to patients’ habitual sleep-wake patterns). However, we conducted follow-up analyses controlling for sleep recording methodology (habitual sleep/wake times versus fixed laboratory time) and results were unchanged. Finally, given that the majority of patients included in these protocols conducted between 1982 and 2001 were treated with tricyclic antidepressants, rather than SSRIs and other current medications, it is possible that findings may differ, in a more contemporaneous pharmacologically treated population. These caveats notwithstanding, this approach of pooling data across individual clinical trials which utilize common assessment tools and standardized treatment protocols offers the powerful opportunity to address research questions that would otherwise be unanswerable by individual clinical trials.
Clinical implications of this research suggest that more aggressive depression treatment, including treatment of sleep disturbances, is warranted in depressed individuals who evidence subjective sleep complaints as well as objective sleep disturbances. Importantly, while the presence of insomnia was nearly ubiquitous in the total sample of depressed patients (73%), only 16–20% of the population had both insomnia and short sleep duration or prolonged sleep latency, respectively. The use of non-invasive, and relatively inexpensive sleep methodologies, such as actigraphy may facilitate the identification of these specific subgroups who may be at increased risk for poor treatment outcome as well as downstream health consequences.