Normalizing sleep quality disturbed by psychiatric polypharmacy: a single patient open trial (SPOT)

There is a growing interest in personalized and preventive medicine initiatives that leverage serious patient engagement, such as those initiated and pursued among participants in the quantified-self movement. However, many of the self-assessments that result are not rooted in good scientific practices, such as exploiting controls, dose escalation strategies, multiple endpoint monitoring, etc. Areas where individual monitoring and health assessments have great potential involve sleep and behavior, as there are a number of very problematic sleep and behavior-related conditions that are hard to treat without personalization. For example, winter depression or seasonal affective disorder (SAD) is a serious, recurrent, atypical depressive disorder impacting millions each year. In order to prevent yearly recurrence antidepressant drugs are used to prophylactically treat SAD. In turn, these antidepressant drugs can affect sleep patterns, further exacerbating the condition. Because of this, possibly unique combinatorial or ‘polypharmaceutical’ interventions involving sleep aids may be prescribed. However, little research into the effects of such polypharmacy on the long-term sleep quality of treated individuals has been pursued. Employing wireless monitoring in a patient-centered study we sought to gain insight into the influence of polypharmacy on sleep patterns and the optimal course of therapy for an individual being treated for SAD with duloxetine (Cymbalta) and temazepam. We analyzed continuous-time sleep data while dosages and combinations of these agents were varied. We found that the administration of Cymbalta led to an exacerbation of the subject’s symptoms in a statistically significant way. We argue that such analyses may be necessary to effectively treat individuals with similar overall clinical manifestations and diagnosis, despite their having a unique set of symptoms, genetic profiles and exposure histories. We also consider the limitations of our study and areas for further research.


Introduction
Winter depression or seasonal affective disorder (SAD) is an atypical depressive disorder that in most cases has onset in fall or winter with remission in spring or summer. It is estimated that approximately 5-10 percent of people in the U.S. (i.e., 10-20 million people) experience varying degrees of SAD in a given year 1 . While full syndromal SAD (frequently dependent on additional external negative stressors) is not reached every year, subsyndromal symptoms can be seen 2 . These symptoms are multiple, and include varying degrees of hypersomnia, carbohydrate-craving and jet-lagged physical and mental states (what is known as "brain fog") resulting in fatigue and irritability. The annual shortening of the photoperiod is believed to be the main factor in SAD onset; however, responses to cold temperatures and epigenetic changes have been documented in seasonal mammals and exhibit evolutionary conservation down to lower forms of life 3-6 , suggesting that many very basic physiologic mechanisms could contribute to SAD. Ultimately, SAD is a complex disease with both chronobiological and neurobiological underpinnings 7-11 , which may include an etiology that for some could even begin in utero 12-16 .
Treating SAD is far from trivial and will require tailoring the treatment to an individual and his or her circumstances, for a whole host of reasons, not the least of which concern both individual and societal expectations regarding work habits, lifestyle, communal conventions surrounding day vs. nighttime activities, and the use of pharmacotherapies to treat conditions affecting behavior. In addition, SAD, and depressive syndromes in general, are known to be accompanied by many co-morbidities and sequelae, including anxiety, detrimental body habitus, anhedonia, and, more importantly, sleep disturbances which may exacerbate any underlying depression as well as the additional associated conditions 2 . Tailored treatments for each and every condition possessed by an individual patient who also has SAD could adversely affect that patient's sleep, thereby creating negative feedback for the SADrelated and other symptoms. Treatment of SAD includes a general recommendation for morning bright light therapy and/or antidepressant treatment which can be somewhat effective in managing symptoms, while melatonin, exercise and negative ion therapy are also suggested. However, a recent critical review of light therapy literature showed that most bright light therapy studies have methodological issues and evidence is not unequivocal 17 . Further, cognitive response to bright light therapy can vary based on genetics 18 . A proper prescription for light therapy requires knowing the dim light melatonin onset (DLMO) of SAD individuals (2/3 are phasedelayed) to determine circadian phase 19 . The same is true for using supplemental melatonin to advance sleep phase, as improper timing and dosing can exacerbate symptoms 19 . Because of the seasonal "on-off" nature of the disorder and difficulty in long-term compliance with bright light therapy (due to eyestrain and lack of individualized prescription), year-round prophylactic treatment with antidepressants may be prescribed.
Treatment for SAD and its sequelae are also compounded for peri-and post-menopausal females -a fact which may be under-appreciated in the primary care setting. The progression to menopause in normal women can result in circadian rhythm, vasomotor, and sleep disturbances and an increased risk for depression, possibly further exacerbating symptoms 20-22 . Therefore, a clinician's choice to potentially increase the dosage of, e.g., a previously effective SSRI antidepressant can in turn exacerbate side effects, such as sleep disturbances. Importantly, sleep apnea is one of the most under-diagnosed conditions in post-menopausal women and is a leading cause of cardiovascular morbidity and mortality [23][24][25][26][27][28][29] . Prescribing sleep medications to aid in depression-related symptoms in peri-or post-menopausal women that may be susceptible, or have, sleep apnea is therefore highly problematic.
The fact that depression and sleep disturbances go hand in hand thus creates even more difficult treatment challenges. For example, ironically, it is known that many first-generation antidepressants exert their effects by, among other things, restoring sleep. Unfortunately, many second-generation antidepressants disrupt sleep. It is now accepted that SSRIs and SNRIs typically used to treat SAD can cause sleep disturbances, both in sleep quality (sleep initiation and maintenance) and sleep architecture (rapid eye movement (REM) and non-REM (NREM) sleep) [30][31][32][33][34][35] . Further, these agents can induce or escalate parasomnias such as periodic leg movements (PLMs) and restless legs syndrome (RLS) 36,37 . These effects on sleep could further lead clinicians to routinely prescribe sleep medications to counter the stimulating effects of antidepressants, as was recommended for insomnia in patients taking fluoxetine [38][39][40] . However, sleep medications can have their own negative impacts on sleep quality and architecture, and are not recommended for

Amendments from Version 1
We have re-titled and extensively revised our manuscript to provide a concise focus on the medication trial aspect of the N-of-1 study, the type of which is properly defined as a single patient open trial (SPOT).
Major sections were shortened, reorganized and amended as follows: • Superfluous patient history was removed from Introduction and Methods • In order to simplify the manuscript, the auxiliary follow up study of mild sleep apnea was moved to the Supplementary Material. Rather than dispose of this work entirely, we chose to continue to make it available for two reasons: • It reflects the reproducibility of the subject's basic no-drug sleep architecture 6 months after the medication trial. • It reveals the impact of a clinical intervention for mild sleep apnea for those who are interested See referee reports REVISED maintenance use. Thus, the resulting polypharmacy used to treat SAD is usually pursued without regard to the timing or dosage of the drugs or concern for drug-drug interactions. This fact, combined with unique patient characteristics such as age, gender, genetic and exposure profile, and co-morbid conditions, can further impact response to any prescribed drug or drug combination and may change over time.
In order to combat these issues, the management of SAD and related psychiatric disorders should, as noted, be pursued in a more patient-specific or 'personalized' manner -something that might not be accomplished at the level of a primary care provider. How such personalization can be achieved generally is an open question given the costs associated with the extra time a clinician might have to spend with a patient to determine an optimal course of therapy, but does suggest a greater number of empirical studies investigating the effects of polypharmacy and the utility of different treatment strategies are needed. In addition, patient-acceptance of the challenges surrounding treatment may motivate selfassessments of the type being pursued by members of the quantified self movement but perhaps in more objective 'N-of-1' clinical trial like settings 41,42 . We describe a study investigating the influence of polypharmacy involving a 58-year-old post-menopausal female who was diagnosed with SAD in 2001. The N-of-1 trial design utilized is known as a "single patient open trial" or SPOT 41 . The SPOT offers an alternative to the typical N-of-1 trial components. A SPOT requires no randomization, no placebo and no blinding and allows limited cross-overs of one or more. The ultimate goals of the study were two-fold: to determine if objective claims about the influence of her treatments on her psychological well-being could be made in a self-assessment-oriented but designed outcome measures study, and whether her medication use correlated with exacerbation of her various symptoms and conditions.
Ultimately, the study leveraged wireless monitoring devices and regression modeling to assess patient sleep quality (e.g., the Zeo Sleep Monitor 43,44 ), and designed a drug removal and dose escalation study to determine drug effects. In the course of the study, a number of important insights were obtained. The study identified a number of statistically significant correlations between medication use and symptomology that led to a number of potential recommendations for future treatments. Although it is important to acknowledge the shortcomings of the study, we feel that such patient-engaged and initiated yet protocol-oriented and designed N-of-1 studies may be the best way to individualize treatments for individuals with multiple mood and sleep-related conditions for which polypharmaceutical interventions are common.

Participant
We studied a post-menopausal 58-year-old female (the 'subject', author VLM) treated for SAD since 2001. The subject was interested in self-monitoring and an N-of-1 study for her sleep disturbances given her lengthy dissatisfaction with available treatment options, lack of insights into her multiple conditions, and a very elaborate and complex treatment history. The subject had a long history of usage of benzodiazepine as a sleep medication while taking antidepressants. The subject loosely qualifies as evening prone or delayed sleep phase disorder according to Basic Language Morningness Scale (BALM) questionnaire, which uses a 6-item scale 45 . In summer 2012, she reported that under prolonged indoor low-light conditions she was susceptible to feeling fatigued, exhibiting seasonal symptomology even in summer months in San Diego. In fall 2012, the subject was taking 60 mg Cymbalta, 30 mg temazepam for sleep, and 100 mg sumatriptan as needed for morning headaches. An N-of-1 (SPOT design) study was pursued to explore how her medications affected her sleep in the context of her diagnosed winter depression (SAD), evening chronotype, delayed sleep phase, restless legs/PLMs and morning headaches.

Ethics
The present study was self-administered by one of the authors (VLM). Therefore, ethical approval from an Institutional Review Board was not sought because the Helsinki Declaration does not apply in this case. Essentially, Cymbalta and temazepam were provided to the subject in pre-specified time periods with pre-specified doses initiated on weekends. Melatonin (Nature Made, 3 mg chocolate melts) was used to attempt to phase-shift the subject as needed to keep a work schedule, but several periods involving different combinations were pursued to explore the influence of melatonin on phase. Consistent with a SPOT design by definition and rationale, the study was pursued without randomization in a real-time, real-life setting, similar to a clinical practice drug de-escalation/withdrawal, and no medication blinding was utilized. In addition, because of the strong effects of the medications on our subject any placebo would have been detected. Similarly, a "no treatment washout period" between treatments was not employed or even feasible. There are several reasons for this, first, not wanting to destroy the continuity of the biological effects; but second, and more importantly, complete Cymbalta withdrawal causes undesirable side-effect symptoms such as "brain-zaps" for several months, the duration of which cannot be predicted. Hence in this case, washouts designed into this type of study would extend the timetable while causing further harms. We accept that this would add carryover and rebound effects at treatment boundaries. As an underlying goal of the study was to eliminate the benzodiazepine temazepam and to determine if any combination of Cymbalta and/or melatonin could normalize our subject's sleep, we took an adaptive approach for which treatment cross-overs were only included in the latter portion of the study. It should also be noted that in designing a study like the one described there are a number of potential confounding variables that inevitably arise especially in any naturalistic, free-living setting assessing sleep quality: a) sleep consolidation could occur as sleep deprivation leads to sleep pressure as week progresses; b) sleeping in and changing sleep patterns on weekends could affect weekday trends; and c) percent time in wake after sleep onset can be increased by PLMs, sleep apnea or other sleep maintenance problems, which could be compounded by medication use.

Sleep analysis.
Each night and morning, the subject manually entered start and stop times into the Zeo sleep monitor iPAD app. The time to REM sleep was manually calculated based on Zeo graphic histogram output showing first REM sleep bar. Percent wake, light, deep and REM sleep and number of awakenings were supplied by the Zeo device. We did not use the Zeo sleep latency parameter "Time to Z" due to the confounding presence of PLMs, which our subject has shown to exhibit upon sleep initiation (clinically validated via videotape). The subject also wore the Actiwatch Spectrum around the clock from April 2013 until August 2013 as well as the PAM-RL ankle sensors nightly from April 2013 to July 2013. Some missing sleep quality data occurred due to days for which the subject was traveling.

General statistical analysis
All analyses were performed using R version 3.

Mathematics
To be more specific, an example model for perstage t was created to follow the simple scheme below, with other variables leveraging similar models: where μ 0 is a y-intercept term, the β terms are regression coefficients, ∈ t is an error term with 0 mean and variance σ 2 . The other terms in the model correspond to the drugs being evaluated and are denoted as follows: Cymbalta 30 mg (cym30); Cymbalta 60 mg (cym60); Melatonin 3 mg (mel3); Cymbalta 30 mg and Melatonin 3 mg (cym30mel3); Cymbalta 30 mg and Melatonin 6 mg (cym30mel6); Cymbalta 60 mg and Melatonin 3 mg (cym60mel3); Cymbalta 60 mg and Melatonin 6 mg (cym60mel6); Cymbalta 60 mg and Temazepam 15 mg (cym60tem15); Cymbalta 60 mg and Temazepam 30 mg (cym60tem30). Significant terms (i.e., p < 0.05 based on t-test of the coefficient value and its standard error) in the model were evaluated in an overall model fit as well as in a step-wise manner. Models were also fit to assess the impact of study design (night in time course) and days of the week (using Sunday as comparator per convention) by including these factors as independent variables in the model. The same analyses were performed for time to REM sleep.

Sleep quality analyses
Sleep data was collected for 188 consecutive nights from December 30, 2012 to July 5, 2013, with 21 nights having missing data (Dataset 1). A description of the 14 trials and the number of nights with complete data are listed in Table 1 (abbreviations: Cymbalta (CYM); temazepam (TEM); melatonin (MEL)). Table 2 gives a descriptive analysis of the sleep parameters used in the study. The mean and standard deviation (SD) for: the number of times per night the subject was awakened (wakes (N)); time to first REM sleep bout in hours (1 st REM (h)); and percentage of time in each sleep stage (wake (%), light (%), deep (%), REM (%)) at each drug dose is shown. The number of days per dose and percent of the total nights are also shown (N days (%)). The dataset was not balanced in the sense that we had different numbers of observations while the subject was on different dosages of a drug.  A clear relationship can be seen between temazepam intake and reduced deep sleep in favor of light sleep ( Figure 1 and      Because of the free-living nature of our study, the subject's polypharmacy and struggle to counter sleep disturbances, a large variability in the data is seen. In addition, "normal" sleep staging typically follows a pattern wherein the first non-REM sleep (    performed as described (see Methods) and data is presented as mean percent for each sleep stage with treatment effects adjusted relative to the intercept. Analyses of percent wake and light sleep met Durbin-Watson test criteria once two outlier nights each were removed. Final model diagnosis showed that all linear regression assumption requirements were satisfied except for the normality condition for percent wake and percent light sleep. Therefore, the Box-Cox procedure and transformations were performed and the models refit. Final models satisfied all diagnostic tests and the transformed mean estimate values (denoted as 'bc') presented in Table 3 were adjusted and back-transformed to give mean percent wake and light sleep.
From Similarly, the impact of an antidepressant such as Cymbalta is expected to show a decrease in REM sleep, mainly through the delay in REM sleep onset (see Table 4). The major impact on wake after sleep onset occurred after the removal of temazepam and during Cymbalta use, indicating a possible sleep maintenance issue. The estimate of the y-intercept (μ 0 ) for the model with wake as the dependent variable suggests that approximately 10.5 percent of the time the subject was in wake without any drug effects (p < 2×10 -16 ). The estimated coefficients for the drug and drug dosage independent variables in the model provide the effect on wake of the drugs. The mean percent wake ranged from 17.8 percent (0.10 (SE: 0.03), p = 0.0034) while the subject was taking 60 mg Cymbalta and 30 mg temazepam to 31.0 percent (0.22 (SE: 0.04), p = 1.3×10 -6 ) while the subject was taking 60 mg Cymbalta and 3 mg melatonin. Interestingly, there is evidence for decreased time classified as wake as the week progresses that might be attributed to a number of things such as increasing sleep pressure during the week, relaxed frame of mind and sleeping in on the weekend. In fact, the decrease in wake to 4.8 percent on Saturday seems to approximately parallel the increase in REM sleep on Saturday (approximately 5 percent) with similar p-values. There was no impact of the night of the study on any of the models. Table 4 shows the univariate analyses of time to REM sleep in hours as a dependent variable. The univariate linear regression model exhibited no serial correlation based on the Durbin-Watson test once two Zeo technical outlier nights were removed (known REML error 43 ). As above, models were also tested for the impact of study design (night in time course) and day of the week. An assessment of the normality and serial correlation among the residuals obtained from the model was performed by Portmanteau test, Durbin-Watson statistic, a standard normality check and ARCH test which showed that all linear regression assumption requirements were satisfied except normality. Therefore, the Box-Cox procedure and transformation was performed, and model refit as above.
The mean estimates presented in Table 4 were adjusted and backtransformed to give the original unit of hours. We used the data to attempt to predict a lower Cymbalta drug dose which might not be expected to interfere with our subject's sleep or perhaps normalize all of the percent sleep stages toward "normal" ranges (i.e., wake 5 percent; light 45-55 percent; deep 20-25 percent; REM 25 percent 46 ) since our subject has increased REM (34 percent) and decreased light (35 percent) sleep if drug effects were accounted for. Table 5 provides the predicted values for 10 mg and 20 mg Cymbalta doses based on the fitted regression models.
We note that even considering the removal of Cymbalta altogether, the percentage of the sleep time our subject was estimated to be in a 'wake' period as detected by the Zeo monitor is high. PLMs that tracked with Cymbalta use did decrease to less than 15 per hour during the study (see Dataset 2) which is considered to be normal and therefore not likely to be a source of confusion for the Zeo monitor since episodes of PLMs may confound time in the wake period. However, micro-arousals and unconscious wakes due to the possible presence of mild sleep apnea in our subject remained a concern and could be reflected in the sleep values we observed. A follow-up study performed to monitor a clinical intervention to correct mild sleep apnea is presented in the Supplementary Material.

Discussion
We have shown that monitoring an individual's response to various drugs used to treat her severe sleep and sleep-related disturbances yielded important and actionable insights. For example, the subject's sleep quality was highly compromised when taking Cymbalta at therapeutic (60 mg) and sub-therapeutic (30 mg) doses and was likely aggravated further by polypharmaceutical interventions she was prescribed. In addition, the subject's other conditions, such as mild sleep apnea, may also have contributed to her sleep disturbances and general physical and psychological health. While sleep disruption is a common side effect of SSRIs and SNRIs, our finding that Cymbalta appears to have exacerbated the subject's condition, is important for personalized care of patients with nuanced conditions. The problems associated with Cymbalta may have been due to the extended release formulation of the drug. It is known that Cymbalta is metabolized by CYP2D6, which has been recently shown to undergo a metabolizer phenotype conversion that cannot be assessed by genetic testing 47 . Drug-induced and particularly co-medication-induced phenoconversion is an increasing problem for personalized medicine 48 . Additionally, temazepam is not a short-acting benzodiazepine drug and can cause hangover effects in the course of a night that could contribute to the phase-delay our subject experienced. In fact, both temazepam and another highly used sleep aid, Ambien, were recently found to be associated with increased morbidity and mortality 49 . Despite the fact our subject was co-morbid for a number of circadian disruptors, her sleep architecture normalized when all drugs were removed. In addition, drug removal unmasked mild sleep apnea, manifesting mainly during an NREM sleep component. The temazepam-Cymbalta combination appears to have induced a removal of deep sleep that actually mimics the shallow sleep architecture seen in depressed patients 50 . Antidepressants are often touted as able to restore deep sleep and delay REM sleep in depression 50 . However, for the subject of focus here (and we suspect many others), the major destruction of her deep sleep occurred when a sleep aid was added to counteract the over-stimulation of the antidepressant.
A number of studies have shown that antidepressants can exacerbate symptoms associated with depression 30-37 . Further, we found that our subject suffered from mild obstructive sleep apnea (OSA) and should probably never have been on sleep medication in the first place. Symptom clusters of poor sleep, migraines, and fatigue should motivate a physician to perform a sleep study. In fact, both in menopausal women and in psychiatric practice where mood and sleep disorders can show bi-directional causation, ordering sleep studies for patients has become the recommended course 51,52 .

Limitations
The drug withdrawal protocol for the subject discussed here ran from December to July. The days were getting longer across the time period (after winter solstice to after summer solstice) so changes Due to the free-living nature of our study, attempts to follow/collect standardized food, exercise and sleep/wake behavior were not maintained, although, attempts to phase-shift to earlier sleep/wake regimens were documented. Applying a SPOT design, there was no randomization, drug placebo, blinding or washouts between trials, but we were able to compare our subject's status to her status at times when no drug was provided in a crossover setting. Abrupt changes in treatment may have contributed some expected and some unexpected noise to the data. For example, temazepam dose decreases would be expected to result in delayed sleep onset, however, changes from Cymbalta 60 mg to Cymbalta 30 mg caused hot-flashes also impacting sleep initiation. For the most part, we collected enough data under each treatment studied (relative to drug or device on/off) to measure effects, including the capture of rebound and recovery effects, and the duration of our individual trial conditions were comparable to what is often seen in sleep literature. As stated in the Methods, our decision was to use a real-time/real-life dose withdrawal and not to use washout periods (the appropriate duration of "washout" would be hard to determine for Cymbalta) to avoid harms. As it was, our Cymbalta dose de-escalation was slower than what is used in clinical practice (Cymbalta 60 mg for 52 days, Cymbalta 30 mg for 84 days, Cymbalta 0 mg for 31 days). We also limited the number of times any one drug combination was provided. Given the number of drugs and the number of doses studied, it would be virtually impossible to accommodate multiple intervals with the same drugs and dosages given. Again, given the strong impact of the drugs used in this case, as evidenced by the variability in the data, only the lower dose trials included cross-overs. This resulted in trials at the beginning of the study having only one measurement, albeit covering periods from 7-25 days each. A similar range in days is seen when the duration of the individual same drug/dose regimen trial replicates are combined.

Conclusions
Many people suffering from circadian and sleep disturbances such as those found in SAD have very unique genetic determinants for their condition, different sets of sleep disturbance sequelae, secondary conditions, and nuanced lifestyles that make it hard to treat them exactly the same way. As a result, more focused attention on what intervention strategy makes the most sense to pursue is required. Such 'personalized' intervention strategies are not trivial to implement since they require an integrated, objective, and often-times completely empirical approach to identify and implement them.
We describe our experience with, and the results of, a comprehensive investigation into the response of a single patient to designed manipulations of her sleep pharmacology. We find that the patient had underlying conditions (e.g., sleep apnea) that were confounded by the use of specific drugs to treat her SAD and that these drugs contributed to, or exacerbated, other issues in the subject's life (e.g., alert time for work, attempts to make up for lack of quality sleep during the week on the weekends, etc.). Ultimately, our study and its results should set a precedent for patient-oriented, yet designed and objective, investigations into the impact of polypharmacy and general drug response in real-world settings.

Consent
Written informed consent for publication of their clinical details and/or clinical images was obtained from the patient.
Author contributions VLM ("the subject") and NJS conceived the idea for the study. VLM designed and implemented experiments, collected and processed all device data, produced graphics and performed all final statistical analyses used in the manuscript, wrote and edited the manuscript. YW performed informatic integration of crossplatform data, initial descriptive, serial correlation and outlier statistical analyses and edited the manuscript. NJS directed all statistical analyses, wrote and the edited manuscript.

Competing interests
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
I confirm that the funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Supplementary sleep apnea intervention analyses
Clinical evaluations suggested that the subject has a deviated septum, a small jaw with substantial retrognathia (overbite), and evidence of clenching and grinding her teeth during sleep. The subject was diagnosed with mild obstructive sleep apnea in August 2013 and a mandibular splint (mouth guard or MG) was fabricated as an intervention. The subject then wore the device as ordered by her physician. After an initial adjustment period, the subject was monitored while the splint was advanced to achieve relief of apnea symptoms (mainly snoring). The monitoring took place for 40 nights from 11-10-2013 through 12-19-2013. As previously described the subject's sleep quality was monitored using the Zeo Monitor. In addition, an Equivital belt (Hidalgo, belt type EQ02) was used to collect the subject's heart rate and R-R interval (beat-to-beat interval) at 15-second intervals (see Methods and Procedures Section below). The unadjusted MG was denoted as MG0x for analysis purposes. Simply inserting a mouth guard creates vertical displacement of the mouth and jaw and also (by design) some horizontal displacement of the jaw. Subsequent adjustments of 4 turns and 6 turns of screws to advance the jaw were denoted as MG4x and MG6x, respectively. Four nights during the time the MG was considered in the analyses had missing data, three nights had no MG wear and 33 nights had the MG worn at settings 0x (13 days), 4x (11 days), and 6x (9 days).
Respiratory sinus arrhythmia is a coupling of the heart rate to follow the breath rate (cardio-pulmonary coupling, CPC). Measures of CPC are usually collected during NREM (light + deep) sleep, which occurs mainly during the first half of the night. In order to accommodate this biological phenomenon and in an attempt to equalize the amount of the data generated by the 15-second Equivital heart rate collection rate, we used an "end-of-night" truncation of 6 am. The collection period used for data analysis was from first sleep bout to last sleep bout before 6 am. This resulted in a minimal change in sleep stage percentages. For example, the 0 mg Cymbalta mean estimates for percent wake (10.5 percent), light (35.4 percent), deep (22.3 percent) and REM (34.2 percent) stages from Table 5 average 15 percent, 33 percent, 23 percent and 30 percent, respectively, when nights are truncated to 6 am. The truncated data shows decreases in percent light and REM sleep which are more prevalent during end-of-night sleep.
Heart rate variability (HRV) is decreased during sleep apnea as the breath is obstructed. An increase in HRV should be seen when obstructive sleep apnea is treated with mandibular advancement. A commonly used measure of HRV is the standard deviation of the R-R interval in milliseconds (ms), also called SDNN. The histograms of R-R interval average (RR-AVG) and the R-R interval standard deviation (RR-STDEV) when grouped by MG adjustments appeared to be roughly normally distributed. One night appeared to be a significant outlier across HRV observations (Thanksgiving night, 11-28-2013, MG setting = 4x) and was removed from all HRV analyses. Excluding this night had a substantial impact on improving model R 2 values, but did not change the overall relationship between HRV and MG settings. We analyzed HRV (RR-STDEV) across the entire night and because HRV might be expected to differ among the different sleep stages, we also analyzed the impact of the MG by sleep stage. Univariate regression analyses were performed as previously described (see Methods and Procedures Section below) and assessments showed that all linear regression assumption requirements were satisfied. Table S1 shows a significant increase in total nightly HRV at the MG0x, MG4x and MG6x advanced settings. Regression coefficients are expressed in the original unit of ms for discussion and treatment effects have been adjusted relative to the intercept. The estimate of the y-intercept (μ 0 ) for the model with HRV as the dependent variable suggests that the subject's HRV was approximately 40ms while wearing no mouth guard. The estimated coefficients for the MG settings as independent variables in the model provide the effect on HRV of the mouth guard. The subject's mean HRV increased to approximately 47ms Interestingly, there was no effect on HRV during REM sleep. This confirms that the effects of MG use we observed are confined to the stages of sleep (NREM) that we felt were expected. None of the models were improved by adding in either night of the study or day of the week.
We also determined the impact that treating sleep apnea had on our subject's sleep quality. Table S2 shows the effect of MG setting on sleep stages. Because of the short time period tested at each MG setting and the non-normality of the percent sleep data, the Box-Cox procedure and transformation was performed, and models refit as above. As above, for discussion, the mean estimates presented in Table S2 are derived from the regression coefficients which have been adjusted and back-transformed to give the original unit of percent.
The estimate of the y-intercept (μ 0 ) for the model with wake as the dependent variable and estimates of the coefficients for MG settings as independent variables suggest that the subject's mean percent wake steadily decreased from approximately 15.6 percent   were truncated at 6 am, which matches the y-intercept estimates (μ 0 ) for the no mouth guard state of the sleep apnea study performed from November-December 2013 (Table S2, 15.6 and 33.3 percent, respectively).
Finally, although we were unable to model percent deep and REM sleep during MG wear, the calculated mean percent deep sleep was 23 percent and the calculated mean percent REM sleep was 29 percent for the 40-night period, which also compares well with the averages calculated from the truncated data (23 percent and 30 percent, respectively). These data indicate a very close agreement in measurements despite an interval of several months and the change in season between the two studies and provides a certain comfort level for the purposes of comparing results from the two interventions.
A clinical sleep study (polysomnography) in January 2014 confirmed the subject to be apnea-free. Ultimately, the final sleep ratios for our subject were "normalized" (wake 2.7 percent; light Heart rate variability analysis. Equivital R-R interval data was collected and after movement artifacts were removed R-R interval nightly averages and R-R interval standard deviations were calculated. Movement artifacts were defined as R-R data spikes < 500 ms and > 1100 ms and the artifact data was imputed by filling in the preceding value. Respiratory sinus arrhythmia is a coupling of the heart rate and breath rate (cardio-pulmonary coupling, CPC). Based on coupled autonomic-respiratory oscillations, "stable" sleep shows high frequency coupling (HFC), "unstable" sleep shows low frequency coupling (LFC), while wake and REM sleep show very low frequency coupling (VLFC) 1 . Therefore, measures of CPC to detect elevated LFC are usually collected during NREM (light + deep) sleep, which occurs mainly during the first half of the night. In order to accommodate this biological phenomenon and in an attempt to equalize the amount of the data generated by the 15-second Equivital heart rate collection rate, we used an "end-of-night" truncation of 6 am. The collection period used for data analysis was from first sleep bout to last sleep bout before 6 am. HRV was defined using the standard deviation of the R-R interval (SDRR), also referred to as the SDNN method (standard deviation of normal-tonormal beats) used by others 2,3 . Actiwatch-defined sleep intervals and Zeo-defined first sleep bouts were used to define beginning of sleep and end of night. This was done to normalize behavior after sleep onset. Given that HRV varies with sleep stage, sleep stage transition and time of night, it was important to define an interval that began with sleep onset and ended at the same time every morning.

Statistics.
For heart rate variability analysis, the data were collected for 40 consecutive nights beginning from November 10, 2013 to December 19, 2013. Four days of missing data were due to Zeo equipment malfunction or a need to wear alternate head devices. We treated missing data in these analyses as missing as random (MAR).
As noted, during this time, the subject wore the mouth guard (MG) for 33 days with settings 0x (13 days), 4x (11 days), and 6x (9 days). The R-R interval data was approximately normal within each MG setting. Variable selection for R-R interval standard deviation (measure of HRV) was performed starting from a full model which included MG setting (0x, 4x, 6x) and manually dropping terms with p-value greater than 0.05. Models were also tested for the impact of study design, days of the week, and model residual diagnosis was assessed as described previously.

Background
The area of the study is highly relevant to sleep medicine and addresses the compounding factors of polypharmacy in a single case study design.
The pathology of SAD is well explained with all of the major aetiologies explored (chronobiological and neurobiology and in-utero environment examined). The individual approach of treating each person's with SAD is outlined -the background discussion is very interesting and relevant to personalised medicine. All current treatment approaches have been discussed so to have the complexities with varying (on-off") nature of SAD.
Difference between first and second generation anti-depressants are discussed in relevance to the impact on sleep quality and sleep architecture. The authors have highlighted this very important fact that SSRIs and SNRs used to treat SAD typically cause sleep issues. This ultimately leads to polypharmacy to treat SAD and sleep related issues.

Methodology
The methodology to determine sleep and activity monitoring is a well validated tool against the gold standard to study sleep -polysomnography. Descriptive statistics are appropriate in this study and report all appropriate information. A univariate regression model was used to determine all important parameters relevant to sleep (including wake, light, deep and REM). This model is appropriate for the study as it provides clear information on the changes in each of the major sleep patterns in response to the trials run for this single case experimental design.

Results of the study
The result of the study highlights the fundamental importance of personalised medicine and should be published. They have proven in this polypharmacy patient that Cymbalta at therapeutic (60mg) and sub-therapeutic (30mg doses) compromised the participants sleep quality. Subsequent trials identified that no drug trial showed the following: Reduced sleep wake cycle Reduced light sleep Increase in deep sleep Increase time in REM sleep. Because of these trials the study identified an underlying sleep apnea issue with this patient -again Increase time in REM sleep. Because of these trials the study identified an underlying sleep apnea issue with this patient -again supporting the methodology used in this study and the benefits of these trials to improve patient health and outcome.

Discussion
The discussion is well balanced and is contrasted to current findings in the literature. Limitations of the study have been well explored. The authors have explored that washout periods were not used in this design, they limited the number of times any one drug combination was provided -Agree with this approach. To-date there are no other studies that could have been included in this section of the paper to support the findings.
The only real criticism was the structure -the HRV data and sleep apnea was at the very end of the article. I wonder if this could be early in the paper -this really highlighted the significance of the n-of-1 trial being able to uncover this finding in the patient. It seems that the way this information is presented could be improved.

If applicable, is the statistical analysis and its interpretation appropriate? Yes
Are all the source data underlying the results available to ensure full reproducibility? Yes

Are the conclusions drawn adequately supported by the results? Yes
No competing interests were disclosed.

Competing Interests:
We have read this submission. We believe that we have an appropriate level of expertise to confirm that it is of an acceptable scientific standard. This version has been cleaned up considerably but is still long. Removing the sleep apnea sections and cleaning up most of the issues between background, methods, results and discussion helps improve the ability to read the document cleanly. Overall the concept of detailed analysis of the effects of various cleaning up most of the issues between background, methods, results and discussion helps improve the ability to read the document cleanly. Overall the concept of detailed analysis of the effects of various medications on sleep for an individual patient is a testament to perseverance and a desire for knowledge. The rigorous approach to a clinical evaluation is impressive.
The results are still hampered by no consideration of carry over effects from one time frame to another. The authors indicate that given the long half-life and perhaps even longer carry-over effect of duloxetine that a pure "washout" would not be practical or possible. This is understood but there are other ways to handle "washout" concerns in N of 1 trials. The most common way is for the data to be removed from analysis during the "cross-over" period. Rarely are pure "placebo" washout periods used. In this report deleting the data from periods of time that a previous drug or level of drug would still be impacting the sleep results is not likely to alter the results from time period to time period. It would likely result in some of the "study" periods being dropped entirely as all or most of the data during one or two periods during the step down of duloxetine would likely be in this window. The section on not being able to conduct a "treatment washout period" should be revised to indicate that a decision was made to not statistically create a washout period in analyzing the data and the reason this approach was not utilized.
The reference to "SPOT" N of 1 trials is not available for review. Though a step-down single testing period is considered a N of 1 in this review from a review of the table of contents, the SPOT study approach was not recognized as a N of 1 approach in the AHRQ contracted series of white papers reviewing the area. Be that as it may, the reasonably rigorous approach to studying medication effects on sleep in a complex clinical situation indicates what is possible with a highly motivated patient and diligent care team. Even without the statistical analysis the graphs of the sleep results are compelling that the changes are a result of changes in medications over time and are useful in making clinical decisions, the main point of the activities.
The term "brain-zap" is not a clinical term and, in fact, does not even appear as a lay term during an internet search. The term should be removed and replaced with a clinical term and referenced. Likewise the term "brain fog" and "jet lagged" are not clinical terms but at least both come up when searched upon. The concepts should be described in clinical terms and the existence of the phenomena should be referenced.
The authors should carefully read through the current version and remove all adjectives and adverbs that do not add scientific value to a sentence. The manuscript has many of these scattered throughout. Some examples are listed below: "Treatment for SAD and its sequelae are also compounded for peri-and post-menopausal femalesa fact which may be under-appreciated in the primary care setting." Either reference this statement as being specific to primary care versus other settings or make the statement more generalized.
"Prescribing sleep medications to aid in depression-related symptoms in peri-or post-menopausal women that may be susceptible,or have, sleep apnea is therefore highly problematic." The term "highly" either needs to be documented or better yet just removed it adds little.
Reference the comment that first generation anti-depressants partially exert their effects through restoring sleep. While the drugs have and are used to help with sleep disturbances at low doses it is not clear that the sleep effects have been clearly related to their effects on depression.
"Thus, the resulting polypharmacy used to treat SAD is usually pursued without regard to the timing or dosage of the drugs or concern for drug-drug interactions." The word "usually" adds nothing and is not documented/referenced. "In order to combat these issues, the management of SAD and related psychiatric disorders should, as noted, be pursued in a more patient-specific or 'personalized' manner -something that might not be accomplished at the level of a primary care provider." In fact the approachs used in this manuscript are not used widely in any setting that this reviewer is aware of -why single out primary care as the issue?
These examples were pulled just from the background section. All sections should be reviewed for similar extraneous words that detract from the clinical message.
No competing interests were disclosed.

Competing Interests:
I have read this submission. I believe that I have an appropriate level of expertise to confirm that it is of an acceptable scientific standard. This manuscript purports to report on an N-of-1 trial to select therapy for a patient with seasonal affective disorder, sleep apnea and polypharmacy. The outcome of the activities was improved sleep quality. The condition, the treatments to be considered and the outcomes all appear to be excellent concepts to submit to N-of-1 trials. The interventional approaches are described in detail. The medication trial is the closest this process comes to a true N-of-1 trial. This said, the entire manuscript appears to be an early draft that requires extensive rework.
This appears to be a complex case study with some quasi-experimental components of the various intervention approaches. The current version of the manuscript mixes the background with the methods section, the methods section with the results section and the discussion covers interventions outside of the medication trials or the manuscript in general. Again, the medication trial is the only part of the process that approaches an N-of-1 trial and should be the focus of the manuscript.
The current manuscript is very long and difficult to follow. The current draft is just under 10,000 words for the primary paper, excluding the abstract, supplemental material and references. This manuscript would be much easier to follow and comprehend if cut to approximately 3500-4000 words, which is already long for a medical article. This will require extensive editing and decisions about what to include and what to exclude. This reviewer cannot provide full editing guidance but the authors need to consult with others that can help craft future versions. that can help craft future versions.
Areas that need to be addressed: The extensive description of the patient can be markedly reduced. Further this component of the manuscript should be in the Methods section as it is essentially a description of the study population. The description should focus on the state of affairs just prior to initiating the medication trial. The remaining background is essentially irrelevant to this case study. The closest part of the therapeutic process that approaches an N-of-1 trial is the medication component. This reviewer recommends focusing on this component of the work if the paper is to be retained as an N-of-1 trail. With this change all the interventions that are discussed prior to or outside of this set of interventions can be dropped and included as the state of the study participant at the start of the trial. The extensive discussion of measurement activities needs to significantly cut and measurement approaches referenced from other literature. The discussion of the measurement approaches is also included in the results section as well as the methods section. Some of the methods section related to the sleep apnea treatment intervention appear to be results in the current draft. This can be solved by dropping the extensive discussion of the sleep apnea diagnosis and intervention entirely as it was not an N-of-1 trial in any sense.
The methods should discuss the N-of-1 approach that was used. The decision to not blind medications should be justified. The cross over pattern selection should discussed. It appears the number of crossovers for each treatment option is limited. This is the primary reason this manuscript appears to be more a case study than a true experimental approach. The number of crossovers should be justified, especially for those medications options that were only studied one time. It appears that the medications were studied primarily in a series of reductions in dosages until the final dosages where there was repeat testing in a back and forth pattern. N-of-1 crossover patterns should be randomized, thus the testing pattern needs to be justified. Given the high variability in sleep quality from night to night the decision to use a limited number of cross-overs seems even more troublesome. The reason for not considering a washout period between treatments needs to be justified.
The discussion of serum markers for major depression disorder is not related to the study methods or results and should be removed. Limitations related to the limited number of crossovers is not discussed. The interference of the mouth guard intervention with the medication trial further complicates the low number of drug crossovers.
This manuscript requires major editing and rewriting prior to being reconsidered.
No competing interests were disclosed.

Competing Interests:
I have read this submission. I believe that I have an appropriate level of expertise to state that I do not consider it to be of an acceptable scientific standard, for reasons outlined above. We would like to thank Dr. Wilson Pace for his insightful review of our manuscript and his many constructive suggestions for improvement. We have amended our manuscript in response and believe it is a much better article as a result.
No competing interests were disclosed. Competing Interests: