Sleep diaries and other subjective measures are essential for the assessment of insomnia

Abstract

It has long been the case that polysomnography (PSG) is neither required nor recommended for the diagnosis or treatment of insomnia (Chesson Jr. et al., 2000; Perlis et al., 2022; Schutte-Rodin et al., 2008; Soldatos et al., 1979). This is because PSG is only dispositive for sleep disorders for which there are occult pathologies (e.g., sleep fragmentation owing to apneic and/or myoclonic events). Several insomnia related biomarkers have been identified including beta/gamma assessed cortical arousal (Perlis et al., 1997; Perlis et al., 2001; Spiegelhalder et al., 2012), objective short sleep duration (Dai et al., 2024; Fernandez-Mendoza et al., 2010; Fernandez-Mendoza et al., 2021), and REM sleep instability (Feige et al., 2023; Riemann et al., 2012). While all have a significant degree of explanatory power, none are pathognomonic for insomnia. These considerations notwithstanding, it is commonly believed that objective measures are preferable. In this editorial, we attempt to clarify why objective measures are not optimal for the assessment of insomnia and why the prospective high density measure of self-reported sleep continuity (sleep diaries) is essential (Grandner & Perlis, 2019; Perlis et al., 2022).

Non-specialists, reviewers, and patients often decry reliance on sleep diaries and argue that since objective measures exist, they should be used (e.g., PSG and/or actigraphy / wearables). After all, as observer-independent measures, objective measures of sleep continuity are believed to be inherently more valid. This claim is likely based on the position that sleep diaries, as subjective measures, are susceptible to measurement errors associated with recall and/or social bias. Recall bias may occur when the estimation of sleep continuity is influenced by memory heuristics (e.g., primacy, saliency, and recency). Social, or contextual, bias may occur when the report of sleep continuity is influenced by demand characteristics (e.g., positive treatment effects from participating in clinical trials). In contrast, PSG is less likely to be affected by such confounds and, as a continuous composite electrophysiological measure, provides for more precise sleep–wake determinations. The former is a qualified statement because there is evidence that placebos produce similar or larger sleep continuity effects on PSG measures compared with sleep diaries (Jiang et al., 2020; Muench et al., 2023; Winkler & Rief, 2015). With respect to the latter (precision), PSG measures may detect sleep or wake bouts that occur in time intervals as brief as 10 s, compared with perceptual judgements that generally occur with a 5–10 min resolution (Bonnet, 1990; Bonnet & Moore, 1982). There is, however, another form of temporal resolution that PSG lacks, the measurement method cannot be repeated with sufficient frequency across nights (repeated to capture night-to-night variability). Given that insomnia severity varies greatly from night-to-night (Buysse et al., 2010; Kay et al., 2013; Perlis et al., 2010; Perlis et al., 2014; Vallieres et al., 2005; Vallieres et al., 2011) (and this is likely why the ICSD-3 & DSM-5 diagnostic criteria for Insomnia Disorder have a frequency standard (American Psychiatric Association, American Psychiatric Association, 2013; Medicine AAoS, 2014)), the ideal measurement strategy must assess initial illness severity and treatment effects using methods that allow for data capture over multiple nights, ideally 7–14 nights per measurement interval (Borba et al., 2020; Buysse et al., 2006; Wohlgemuth et al., 1999). Such sampling necessarily allows for a better measure of central tendency.

When single-night PSG and sleep diary summary data are compared (e.g., sleep latency [SL], wake after sleep onset [WASO], early morning awakenings [EMA], and total sleep time [TST]), the measures are often discordant (Benz et al., 2023; Carskadon et al., 1976; Edinger & Fins, 1995; Harvey & Tang, 2012; Perlis et al., 1997). This likely occurs for several reasons; some of which are related to the limitations of subjective measures (as above). Other reasons are related to methodological issues with the objective measures (EEG), including: the source potential of EEG activity (derives from slow vs. action potentials (Brienza & Mecarelli, 2019)); the traditional electroencephalographic filter settings for PSGs (highlights sleep and not wake frequencies (Perlis et al., 1997; Perlis et al., 2001; Spiegelhalder et al., 2012)); the 30 s epoch used for PSG scoring (likely exceeds the limits of the human perception of sleep and wakefulness (Bonnet, 1990; Bonnet & Moore, 1982)); and the use of precision scoring (identifies short bouts of sleep that occur during extended wakefulness that are likely perceived as wakefulness (Knab & Engel, 1988)). These methodological considerations may account for a large proportion of the variance regarding why patients perceive more nocturnal wakefulness and less total sleep time than is detected by PSG (Perlis et al., 1997; Perlis et al., 2001; Spiegelhalder et al., 2012). Further, they likely also account for why patients perceive more treatment-related improvements on subjective measures (Holbrook et al., 2000; Mendelson, 1993).

If repeated measures are essential, why not use actigraphy (or like commercial wearables) to provide objective measures of sleep continuity on a night-to-night basis? While it is true that motion/activity detection devices allow for the repeated measure of sleep continuity over long time intervals (and do so with a minimum of effort by the subject), these positive attributes don't make wearables a more valid measure of sleep continuity (just more reliable). These devices do not assess the aspects of sleep experienced as insomnia. Most wearable devices primarily use movement-based sleep–wake detection where the absence of activity (extended immobility) is used as a proxy measure of sleep. This measurement approach reliably allows for state discrimination in non-insomnia populations (de Souza et al., 2003; Sadeh, 2011; Sadeh & Acebo, 2002). In patients with insomnia, where the individual is inactive but awake, wearables tend to underestimate wakefulness (Lichstein et al., 2006; Paquet et al., 2007; Vallieres & Morin, 2003).

Whatever the objective measurement strategy (PSG or wearables), they may be inherently incomplete (i.e., less valid) because of the clinical diagnostic criteria that insomnia severity be defined and assessed idiographically (i.e., what it means to take too long to fall asleep and/or what it means to be awake for too long during the night is defined by the individual) (American Psychiatric Association, 2013; Medicine AAoS, 2014). Even if these were objectively and/or quantitatively assessed (as is often done in clinical research), the definition of insomnia still requires that the patient experience sleep continuity disturbance as a concern or as a cause of impaired daytime function. At present, there is no objective measure for the patient's sense of “dis-ease” and there is no amalgam of measures that provide for a proxy of the individual's sense of suffering. This being the case, the only way to assess problem perception is via self-report. Doing so is necessary for the assessment of illness severity and treatment response. In the case of treatment response, “the simple reliance on objective measures risks the detection of objective improvement in the absence of perceived improvement; and this is tantamount to no improvement” (Grandner & Perlis, 2019; Perlis et al., 2022). Within the sleep community, such a proposition may be the opposite of conventional wisdom (i.e., subjective improvement in the absence of objective improvement is little more than a placebo effect). Perhaps this point can be better illustrated in terms of the clinical problem of chronic pain. If one assumes that an objective measure of nociception existed, treatment-related change on such measures without a diminution in the experience of self-reported pain intensity would be considered ineffective therapy (by both the patient and the clinician) (Grandner & Perlis, 2019; Perlis et al., 2022). Given the centrality of the experience of insomnia (like the experience of pain), objective measures can only be used as a primary measure when they can be shown to be highly concordant with sleep diaries. This state of affairs is likely to persist until a pathophysiological measure is found that robustly corresponds to the presence or absence of insomnia; perceived insomnia severity; and incidence/intensity of daytime sequelae (i.e., until a measure is found that is pathognomonic for insomnia). Until such a measure is found, it is incumbent on us to make the prospective high density sampling of self-reported sleep continuity (i.e., sleep diaries) more reliable. This will require finding ways to enhance adherence by making the data capture process faster and more engaging.

The arguments above are provided to address two issues. First, to call into question the claim that PSG or wearables are inherently more valid. Second, to substantiate the claim that sleep diaries are the optimal measure of sleep continuity disturbance precisely because they are prospective and high-density measures. That is, measures that allow for the repeated assessment of the patient's sense of illness severity and do so in a manner that uses common sense and continuous metrics (minutes and hours) that are specific to patient complaint (i.e., initial, middle, and late insomnia) (Grandner & Perlis, 2019).

As indicated above, the measurement of sleep continuity alone is not sufficient for the assessment of illness severity and/or treatment-related change. To accomplish this, assessment must also take into account the patient's sense of sleep dissatisfaction and/or the incidence and severity of daytime sleep-related impairment. These components of insomnia (which not only quantify the level of suffering but speak to whether or not sleep need is met) are generally not assessed with sleep diaries. Instead, these components of insomnia are evaluated with dedicated retrospective instruments such as the Insomnia Severity Index (ISI) (Bastien et al., 2001; Morin et al., 2011). The ISI is often recommended as the single best outcome measure for insomnia because it concurrently assesses sleep continuity disturbance, sleep dissatisfaction, and sleep-related impairment (Ji et al., 2019). The main limitation of the ISI, and related measures, is its retrospective frame and qualitative descriptors of severity. With respect to the ISI's scaling, its descriptors are not arrayed in terms of natural units (minutes-to-hours of initial, middle, or late insomnia per typical night or days per week above a given threshold) but instead are scaled in qualitative terms (“none”, “mild”, “moderate”, “severe”, “very severe”). Such response sets are likely to increase individual error variance (i.e., one person may experience a symptom as “moderate” whereas another may experience the same symptom intensity as “severe”). While such a qualitative difference may speak to whether or not sleep need is met, such a relative assessment is not needed given the other items on the instrument (as below). The other potential limitation of the ISI derives from another of its major strengths; its implicit two factor structure (1st three items pertain to sleep continuity disturbance and 2nd four items pertain to problem endorsements). Given this, and its sum scoring threshold (≥ 15 is moderate insomnia), it is possible to be identified as having insomnia with minimal sleep continuity disturbance or as not having insomnia despite having substantial sleep continuity disturbance; the latter potentially being characteristic of older adults.

In sum, insomnia (like pain) is a psychophysiological condition. Unlike pain, objective measures of illness severity and frequency are available. The existence of such measures should not preempt (or render as obsolete) the use of subjective measures. Quite the contrary, sleep diaries and retrospective measures of sleep related impairment are essential for the assessment of perceived illness severity and recovery. Objective measures, while inherently more reliable, are not more valid. Objective measures should be viewed as complementary assays of the potential causes, correlates, and consequences of insomnia. Ultimately, insomnia is best assayed using a multi-method multi-trait approach (MMT). The minimum MMT assessment should include sleep diaries (for the quantitative assessment of sleep continuity) and the ISI (for the assessment of perceived consequence). Better still would be to include measures from both the subjective and objective domains; measures that also embrace other health-and-performance related data (Morin, 2003). The emergent question here will be how to gather prospective high frequency sampled data in a manner that allows for simultaneous data acquisition or temporal alignment of differently sourced data. Ideally, this will occur with digital platforms that make good use of APIs (Application Programming Interfaces) so that multiply sampled sleep, performance, and health data are integrated into relational databases that are temporally synchronised. As the MMT approach becomes more the norm, it is likely that additional discrepancies and discordances will be observed and need to be reconciled (e.g., when various measures of sleep continuity disturbance don't align or when illness severity and daytime dysfunction are not concordant) (Boyle et al., 2022). Rather than view such discrepancies and discordances as examples of the inferiority of one measure as opposed to another, it may be best to view these as opportunities to understand the strengths and limitations of each method, and as opportunities to ask and answer the question “what accounts for the differences between measures?” (Boyle et al., 2022).

Michael Perlis: Conceptualization; writing – original draft; writing – review and editing. Michael Grandner: Conceptualization; writing – original draft; writing – review and editing. Donn Posner: Conceptualization; writing – review and editing. Kai Spiegelhalder: Conceptualization; writing – review and editing; writing – original draft. Dieter Riemann: Conceptualization; writing – review and editing.

MLP: R01AG054521, K24AG055602, & Axsome Therapeutics.

No conflicts of interest are reported by any of the authors.