**3.1 Self-report sleep data**

*Psychopathology - An International and Interdisciplinary Perspective*

four studies. When training was complete, studies were randomly assigned to one of the four undergraduate judges. The graduate student acted as second judge and independently coded all studies. Regular meetings were held to review discrepancies and reach consensus. In total, 29 of the 31 studies (93.5%) included in the meta-analysis were independently rated by two judges. The percentages of interrater agreement were between 70 and 100% for the majority of coded variables (see details in **Table 1**). Lower agreement rates were observed on variables related to NPA, a finding that can be explained by a lack of precision in reports of NPA

Results were analyzed using the comprehensive meta-analysis software, version 3. Since the studies included in this review varied widely, a random effects model was used. Many of the studies had small sample sizes, in particular polysomnographic studies. Hedges' g was consequently chosen for effect size because of its

each analysis of significant results including three studies or more, Orwin's fail-safe N was used to investigate publication bias. In these cases, a funnel plot was visually

A total of 1229 articles were screened for eligibility (**Figure 1**). Of the 1229,80 initially met inclusion criteria. Forty-nine were subsequently excluded. A total of 31 studies were selected for review. The majority of studies with self-report or polysomnographic data reported no significant differences in age or gender ratio

were used to assess heterogeneity. For

**100**

**Figure 1.**

prevalence.

examined.

**3. Results**

**2.5 Data analysis**

correction for small samples [32]. Q and I2

*Study selection flow chart.*

Of the 31 selected articles, seven reported self-report data. All seven studies used the PSQI. The component scales of subjective sleep quality, sleep latency, sleep duration, sleep efficiency, sleep disturbances, daytime dysfunction, and the PSQI global score were analyzed. The sixth PSQI scale (medication use) was not analyzed because the variable was outside the scope of the present study. Pooled PD group size ranged from 47 to 182 participants (average = 94 participants). Pooled control groups ranged from 15,151 to 15,238 participants (average = 15,163 participants) (**Table 2**). The wide difference between average group size in PD and control groups was due to the presence of one study whose control group was 15,117 participants. When this study was removed from calculations of the average sample size, average control group size was 46 participants.

Analyses revealed that patients with PD had significantly greater scores than healthy controls for the PSQI scales of sleep quality, sleep latency, sleep disturbances, daily dysfunction, and global score. This means that patients with PD reported worse sleep quality, longer sleep latency, more sleep disturbances, and more daily dysfunction. All effect sizes were large as per Cohen's criteria [44]. There were no significant differences between groups on PSQI sleep efficiency and sleep duration scales.


*Note: PSQI = Pittsburgh sleep quality index; — = insufficient number of studies to report heterogeneity;* 

*CI = confidence interval.*

*\*= Significant differences between groups, p < 0.05.*

*\*\*= Significant heterogeneity, p < 0.05.*

#### **Table 2.**

*Meta-analysis results for self-report data.*

To evaluate the consistency of data across studies, heterogeneity tests were performed on data for PSQI global score. The results indicated the presence of significant heterogeneity. Estimated variance in true effects (I2 ) was 92.8%. The small number of effect sizes precluded subgroup or meta-regression analyses that may have explained the heterogeneity of results.

Among subjective sleep variables, the impact of publication bias was assessed for PSQI global score. Results of Orwin's fail-safe N indicated that 48 unpublished studies with no effect would be required to lower the combined effect for PSQI global score to 0.2 (criterion for trivial effect). Moreover, the analysis of the funnel plot revealed an uneven distribution of studies, the majority of which were below the mean Hedges' g value. That is, both Orwin's fail-safe N and funnel plot analysis suggest the possibility of publication bias due to a "file drawer effect" (i.e., fewer studies with unsignificant results are published). The results must therefore be interpreted with caution.

#### **3.2 Polysomnographic sleep data**

Of the 31 included studies, 15 reported polysomnographic or actigraphic data. Meta-analysis was performed with 18 variables: sleep latency, sleep efficiency, number of awakenings, total sleep time, total wake time, REM sleep characteristics (duration, density, percentage, latency, number of REM periods), slow-wave sleep characteristics (duration, percentage), duration of stage 1 and stage 2 sleep, and percentage of stages 1–4 sleep. Pooled samples sizes varied between 19 and 209 participants for PD groups (average = 115 participants) and between 19 and 164 participants for control groups (average = 86 participants) (**Table 3**).

Significant differences between PD and control groups were demonstrated for four variables. PD participants had longer sleep latency, lower sleep efficiency, and shorter total sleep time than healthy controls. Duration of stage 2 sleep was also shorter. Analyses for the 14 other sleep variables did not show statistical significance.

To evaluate the consistency of data across studies, heterogeneity tests were performed on data for sleep latency, sleep efficiency, and total sleep time. The results indicated the presence of significant heterogeneity for sleep efficiency and for total sleep time. For the former, estimated variance in true effects (I2 ) was 52%. For the latter, estimated variance in true effects (I2 ) was 68%. There was no significant heterogeneity among sleep latency effect sizes. The small number of effect sizes precluded subgroup or meta-regression analyses that may have explained the heterogeneity of results.

We assessed the possibility of publication bias for the following variables: sleep latency, sleep efficiency, and total sleep time. For sleep latency, Orwin's fail-safe N was 31 (criterion for trivial Hedges' g = 0.2). This statistic indicates that it would take 31 studies with no effect to decrease the effect size to 0.2 or lower. Similar analysis yielded Orwin's fail-safe N of 26 for sleep efficiency and 24 for total sleep time. The results indicated that total effect size for each variable could be affected by publication bias. However, analysis of funnel plots, which are relatively wellbalanced, indicated a lower probability of publication bias. When Orwin's fail-safe N and funnel plot analysis are combined, the possibility of publication bias cannot be excluded.

#### **3.3 Prevalence of NPA**

Thirteen studies were included in the analysis of the prevalence of NPA in patients with PD. Data was classified into three distinct categories according to

**103**

reported NPA frequency: at least one lifetime NPA, one or more NPA in the past month, and two NPA per month or per 2 months with apprehensiveness about possible future NPA. For the latter criterion, the intensity of apprehensiveness about NPA had to be a minimum of four on a scale of 1–8. Finally, both NPA and

*in effect size; PD = panic disorder; REM sleep = rapid eye movement sleep; slow-wave sleep = stages 3 and 4 sleep;* 

*A Meta-Analysis of Sleep Disturbances in Panic Disorder DOI: http://dx.doi.org/10.5772/intechopen.86306*

> **Confidence interval (95%)**

> > −0.43]

−0.35]

0.50]

0.32]

0.20]

0.24]

0.57]

0.28]

−0.12]

0.47]

*Meta-analysis results for polysomnographic and actigraphic data.*

*— = insufficient number of studies to report heterogeneity; CI = confidence interval.*

**Standard error**

Sleep latency 0.81\* [0.58, 1.04] 0.12 0.01 7.95 0.00 10 168 147

**Variance Q I**

**<sup>2</sup> Number of studies**

0.18 0.03 16.20\*\* 50.62 9 186 118

0.17 0.03 2.35 0.00 4 80 66

0.20 0.04 31.06\*\* 67.80 11 209 164

0.40 0.16 27.02\*\* 81.49 6 88 68

0.32 0.10 4.15 51.76 4 65 51

0.19 0.03 1.73 0.00 4 64 52

0.11 0.01 6.98 0.00 10 181 134

0.25 0.06 33.48\*\* 76.10 9 187 140

0.32 0.11 3.60 44.38 3 40 30

0.28 0.08 — — 2 27 23

0.18 0.03 22.53\*\* 60.05 10 174 132

0.31 0.10 — — 2 19 19

0.28 0.08 — — 2 27 23

0.12 0.01 3.71 0.00 9 164 137

0.25 0.06 39.89\*\* 77.44 10 190 153

0.35 0.12 10.43\*\* 71.24 4 97 54

0.40 0.16 13.28\*\* 77.41 4 97 54

 *= proportion of variance due to real differences* 

**PD group sample size**

**Control group sample size**

**Variable Effect** 

Sleep efficiency

Number of awakenings

Total sleep time

Total wake time

REM sleep duration

REM sleep density

Number of REM periods

Slow-wave sleep duration

% slow-wave sleep

Stage 1 duration

Stage 2 duration

% stage 1 sleep

% stage 2 sleep

% stage 3 sleep

% stage 4 sleep

**Table 3.**

**size (Hedges' g)**

−0.78\* [−1.12,

0.07 [−0.25, 0.40]

−0.74\* [−1.13,

0.58 [−0.20, 1.36]

−0.12 [−0.75,

−0.04 [−0.41,

0.04 [−0.60, 0.67]

0.03 [−0.51,

−0.07 [−0.42,

0.25 [−0.36, 0.87]

−0.68\* [−1.23,

0.13 [−0.11, 0.36]

0.19 [−0.30, 0.67]

0.31 [−0.38, 1.00]

−0.31 [−1.10,

*\*= Significant differences between groups, p < 0.05.*

*\*\*= Significant heterogeneity, p < 0.05.*

*Note: Q = weighted sum of squares, indicates total dispersion; I2*

% REM sleep −0.01 [−0.22,

REM latency −0.25 [−0.73,


#### *A Meta-Analysis of Sleep Disturbances in Panic Disorder DOI: http://dx.doi.org/10.5772/intechopen.86306*

*Psychopathology - An International and Interdisciplinary Perspective*

significant heterogeneity. Estimated variance in true effects (I2

may have explained the heterogeneity of results.

interpreted with caution.

significance.

heterogeneity of results.

**3.2 Polysomnographic sleep data**

To evaluate the consistency of data across studies, heterogeneity tests were performed on data for PSQI global score. The results indicated the presence of

small number of effect sizes precluded subgroup or meta-regression analyses that

Among subjective sleep variables, the impact of publication bias was assessed for PSQI global score. Results of Orwin's fail-safe N indicated that 48 unpublished studies with no effect would be required to lower the combined effect for PSQI global score to 0.2 (criterion for trivial effect). Moreover, the analysis of the funnel plot revealed an uneven distribution of studies, the majority of which were below the mean Hedges' g value. That is, both Orwin's fail-safe N and funnel plot analysis suggest the possibility of publication bias due to a "file drawer effect" (i.e., fewer studies with unsignificant results are published). The results must therefore be

Of the 31 included studies, 15 reported polysomnographic or actigraphic data. Meta-analysis was performed with 18 variables: sleep latency, sleep efficiency, number of awakenings, total sleep time, total wake time, REM sleep characteristics (duration, density, percentage, latency, number of REM periods), slow-wave sleep characteristics (duration, percentage), duration of stage 1 and stage 2 sleep, and percentage of stages 1–4 sleep. Pooled samples sizes varied between 19 and 209 participants for PD groups (average = 115 participants) and between 19 and 164

Significant differences between PD and control groups were demonstrated for four variables. PD participants had longer sleep latency, lower sleep efficiency, and shorter total sleep time than healthy controls. Duration of stage 2 sleep was also shorter. Analyses for the 14 other sleep variables did not show statistical

To evaluate the consistency of data across studies, heterogeneity tests were performed on data for sleep latency, sleep efficiency, and total sleep time. The results indicated the presence of significant heterogeneity for sleep efficiency and

nificant heterogeneity among sleep latency effect sizes. The small number of effect sizes precluded subgroup or meta-regression analyses that may have explained the

We assessed the possibility of publication bias for the following variables: sleep latency, sleep efficiency, and total sleep time. For sleep latency, Orwin's fail-safe N was 31 (criterion for trivial Hedges' g = 0.2). This statistic indicates that it would take 31 studies with no effect to decrease the effect size to 0.2 or lower. Similar analysis yielded Orwin's fail-safe N of 26 for sleep efficiency and 24 for total sleep time. The results indicated that total effect size for each variable could be affected by publication bias. However, analysis of funnel plots, which are relatively wellbalanced, indicated a lower probability of publication bias. When Orwin's fail-safe N and funnel plot analysis are combined, the possibility of publication bias cannot

Thirteen studies were included in the analysis of the prevalence of NPA in patients with PD. Data was classified into three distinct categories according to

participants for control groups (average = 86 participants) (**Table 3**).

for total sleep time. For the former, estimated variance in true effects (I2

For the latter, estimated variance in true effects (I2

) was 92.8%. The

) was 52%.

) was 68%. There was no sig-

**102**

be excluded.

**3.3 Prevalence of NPA**

*Note: Q = weighted sum of squares, indicates total dispersion; I2 = proportion of variance due to real differences in effect size; PD = panic disorder; REM sleep = rapid eye movement sleep; slow-wave sleep = stages 3 and 4 sleep; — = insufficient number of studies to report heterogeneity; CI = confidence interval.*

*\*= Significant differences between groups, p < 0.05.*

*\*\*= Significant heterogeneity, p < 0.05.*

#### **Table 3.**

*Meta-analysis results for polysomnographic and actigraphic data.*

reported NPA frequency: at least one lifetime NPA, one or more NPA in the past month, and two NPA per month or per 2 months with apprehensiveness about possible future NPA. For the latter criterion, the intensity of apprehensiveness about NPA had to be a minimum of four on a scale of 1–8. Finally, both NPA and


*Note: Q = weighted sum of squares, indicates total dispersion; I2 = proportion of variance due to real differences in effect size; PD = panic disorder; NPA = nocturnal panic attack; — = insufficient number of studies to report heterogeneity; CI = confidence interval.*

*\*= Significant heterogeneity, p < 0.05.*

#### **Table 4.**

*Meta-analysis results for NPA prevalence.*

apprehensiveness had to be present for at least the prior 6 months. A meta-analysis was performed for each of the three categories, with frequency of NPA in patients with PD and PD group sample size as input data. Results indicated that, among the pooled sample of patients with PD, an average of 52.1% (95% CI [46.4, 57.7]) reported at least one lifetime NPA (**Table 4**). The heterogeneity test revealed significant heterogeneity between studies and indicated that 73% of observed variance was attributed to true effects. The prevalence rate of one or more NPA in the past month was 27.0% (95% CI [17.9, 38.6]), and the heterogeneity test was significant. Sixty-five percent of observed variance was attributed to true effects. For recurrent NPA (2/month or/2 months with apprehensiveness about possible future NPA, intensity of apprehensiveness of minimally four on a scale of 1–8.), the mean prevalence rate was 40.9% (95% CI [18.1, 68.5]). As analyses included only two studies, heterogeneity was not calculated.

#### **4. Discussion**

This study was designed to use meta-analytic methodology to draw a detailed portrait of sleep disturbances in PD. More specifically, the primary objective was to compare sleep in this population to sleep in healthy controls. Results from subjective and objective data analysis confirmed the hypothesis that sleep quality in the former group is significantly poorer than in the latter group. Furthermore, in comparison to controls, patients with PD take longer to fall asleep and have more sleep disturbances and more difficulty with daytime functioning secondary to sleep problems. Analysis of objective sleep data revealed differences between patients with PD and healthy controls in sleep continuity parameters: patients with PD take longer to fall asleep, have a shorter sleep duration, and demonstrate poorer sleep efficiency. For the majority of sleep architecture parameters, no differences were noted between patients with PD and control participants, with the exception of stage 2 duration, which was shorter in patients with PD.

#### **4.1 Self-report data**

To date, some literature reviews have explored subjective sleep complaints in PD [3, 38–40]. They reported conclusions that are consistent with the data reported here, indicating that patients with PD report significant subjective sleep alterations. However, most reviews did not detail the nature of these complaints. For example, Mellman [39] reported that there was subjective insomnia in patients with

**105**

**4.2 Objective data**

rather than sleep maintenance.

*A Meta-Analysis of Sleep Disturbances in Panic Disorder DOI: http://dx.doi.org/10.5772/intechopen.86306*

PD, but without further specifying if there was a problem with sleep onset, sleep maintenance, or early awakenings. Only one recent systematic literature review has provided greater precision on subjective sleep complaints by reporting the results of nine previous studies in a structured manner [3]. This study reports a wide range of measures such as the sleep-wake experience list [45], the PSQI [18, 21, 46, 47], and the Goldberg depression and anxiety scales [48]. This unites useful and diverse information about the sleep of patients with PD. However, equivalence of measures and results synthesis remains difficult to judge. The statistical procedures of metaanalysis that we used allowed us to carry out the said synthesis and provided easily

The elevated level of heterogeneity in PSQI global scores constitutes one critical point for consideration in our analysis of subjective sleep data. Variation in effect sizes is significant, and the majority of variance (92.8%) is attributable to differences in true effects. The observed elevated level of heterogeneity may be attributable to hidden covariates that moderate the differences between patients with PD and healthy controls. Previous research has identified subgroups of patients with PD in which insomnia might be more prevalent, including patients reporting NPA, depression [45], and greater anxiety sensitivity [18]. Effect sizes reported in each individual study may vary according to whether participants from these subgroups were included. Unfortunately, the number of studies reporting PSQI global scores was too low to permit the investigation of the impact of such variables via metaregression [32]. Further research addressing the impact of covariates such as NPA, depression, and anxiety sensitivity on sleep quality in PD patients is warranted.

Among objective sleep variables, previous literature reviews reported impairments in sleep continuity parameters. Reports of sleep onset latency, sleep efficiency, and total sleep time alterations were the most cited and robust [3, 39, 40]. Some authors also mentioned that patients with PD have difficulty maintaining sleep [39], with higher percentage of wake time [40]. Combined effect sizes from our findings indeed confirm that patients with PD have shorter sleep, take longer to fall asleep, and have poorer sleep efficiency. However, they do not confirm previous findings regarding sleep maintenance, since total wake time and number of awakenings were similar in PD and control groups. Based on these results, the nature of sleep difficulties in PD seems to be restricted to sleep onset and early awakenings,

Previous reviews highlighted the existing inconsistencies in sleep architecture data. Overall, they report that there is as much evidence showing that the sleep architecture of patients with PD and of healthy controls is different as there is evidence that they are not [3, 40]. The narrative nature of previous reviews limited the conclusions that could be drawn from such data. Given this limit, the present study used meta-analysis to synthesize conflicting data. Results showed no difference between the sleep of patients with PD and the sleep of healthy control participants on the percentages of stage 1 and stage 2 sleep, of slow-wave sleep, and of REM sleep. Also, there were no differences between groups for the number of REM periods, REM sleep density and duration, stages 1 and 2 sleep duration, slow-wave sleep duration, and percentages of stage 3 and of stage 4 sleep. However, the latter results were less reliable due to the small number of studies that were included (two to four studies for each variable). Therefore, confirmation of the results is needed. In sum, there is reliable evidence that, in comparison to healthy controls, patients with PD demonstrate alterations in objective sleep latency, sleep efficiency, and total sleep time. Also, it is plausible that percentages of REM sleep, of delta sleep, and of

understandable summary indicators (size effects) of the results.

#### *A Meta-Analysis of Sleep Disturbances in Panic Disorder DOI: http://dx.doi.org/10.5772/intechopen.86306*

*Psychopathology - An International and Interdisciplinary Perspective*

**(prevalence %)**

*Note: Q = weighted sum of squares, indicates total dispersion; I2*

**Frequency criteria Event rate** 

2 NPA/month or /2 months with apprehensiveness 4/8, lasting at least 6 months

**Table 4.**

*heterogeneity; CI = confidence interval. \*= Significant heterogeneity, p < 0.05.*

*Meta-analysis results for NPA prevalence.*

two studies, heterogeneity was not calculated.

which was shorter in patients with PD.

**4.1 Self-report data**

**4. Discussion**

apprehensiveness had to be present for at least the prior 6 months. A meta-analysis was performed for each of the three categories, with frequency of NPA in patients with PD and PD group sample size as input data. Results indicated that, among the pooled sample of patients with PD, an average of 52.1% (95% CI [46.4, 57.7]) reported at least one lifetime NPA (**Table 4**). The heterogeneity test revealed significant heterogeneity between studies and indicated that 73% of observed variance was attributed to true effects. The prevalence rate of one or more NPA in the past month was 27.0% (95% CI [17.9, 38.6]), and the heterogeneity test was significant. Sixty-five percent of observed variance was attributed to true effects. For recurrent NPA (2/month or/2 months with apprehensiveness about possible future NPA, intensity of apprehensiveness of minimally four on a scale of 1–8.), the mean prevalence rate was 40.9% (95% CI [18.1, 68.5]). As analyses included only

**Confidence interval (95%)**

1 or more lifetime NPA 52.1 [46.4, 57.7] 36.51\* 75.34 10 1647 1 NPA in the last month 27.0 [17.9, 38.6] 8.47\* 64.59 4 224

*in effect size; PD = panic disorder; NPA = nocturnal panic attack; — = insufficient number of studies to report* 

**Q I**

 *= proportion of variance due to real differences* 

40.9 [18.1, 68.5] — — 2 221

**<sup>2</sup> Number of studies**

**Sample size**

This study was designed to use meta-analytic methodology to draw a detailed portrait of sleep disturbances in PD. More specifically, the primary objective was to compare sleep in this population to sleep in healthy controls. Results from subjective and objective data analysis confirmed the hypothesis that sleep quality in the former group is significantly poorer than in the latter group. Furthermore, in comparison to controls, patients with PD take longer to fall asleep and have more sleep disturbances and more difficulty with daytime functioning secondary to sleep problems. Analysis of objective sleep data revealed differences between patients with PD and healthy controls in sleep continuity parameters: patients with PD take longer to fall asleep, have a shorter sleep duration, and demonstrate poorer sleep efficiency. For the majority of sleep architecture parameters, no differences were noted between patients with PD and control participants, with the exception of stage 2 duration,

To date, some literature reviews have explored subjective sleep complaints in PD [3, 38–40]. They reported conclusions that are consistent with the data reported here, indicating that patients with PD report significant subjective sleep alterations. However, most reviews did not detail the nature of these complaints. For example, Mellman [39] reported that there was subjective insomnia in patients with

**104**

PD, but without further specifying if there was a problem with sleep onset, sleep maintenance, or early awakenings. Only one recent systematic literature review has provided greater precision on subjective sleep complaints by reporting the results of nine previous studies in a structured manner [3]. This study reports a wide range of measures such as the sleep-wake experience list [45], the PSQI [18, 21, 46, 47], and the Goldberg depression and anxiety scales [48]. This unites useful and diverse information about the sleep of patients with PD. However, equivalence of measures and results synthesis remains difficult to judge. The statistical procedures of metaanalysis that we used allowed us to carry out the said synthesis and provided easily understandable summary indicators (size effects) of the results.

The elevated level of heterogeneity in PSQI global scores constitutes one critical point for consideration in our analysis of subjective sleep data. Variation in effect sizes is significant, and the majority of variance (92.8%) is attributable to differences in true effects. The observed elevated level of heterogeneity may be attributable to hidden covariates that moderate the differences between patients with PD and healthy controls. Previous research has identified subgroups of patients with PD in which insomnia might be more prevalent, including patients reporting NPA, depression [45], and greater anxiety sensitivity [18]. Effect sizes reported in each individual study may vary according to whether participants from these subgroups were included. Unfortunately, the number of studies reporting PSQI global scores was too low to permit the investigation of the impact of such variables via metaregression [32]. Further research addressing the impact of covariates such as NPA, depression, and anxiety sensitivity on sleep quality in PD patients is warranted.

#### **4.2 Objective data**

Among objective sleep variables, previous literature reviews reported impairments in sleep continuity parameters. Reports of sleep onset latency, sleep efficiency, and total sleep time alterations were the most cited and robust [3, 39, 40]. Some authors also mentioned that patients with PD have difficulty maintaining sleep [39], with higher percentage of wake time [40]. Combined effect sizes from our findings indeed confirm that patients with PD have shorter sleep, take longer to fall asleep, and have poorer sleep efficiency. However, they do not confirm previous findings regarding sleep maintenance, since total wake time and number of awakenings were similar in PD and control groups. Based on these results, the nature of sleep difficulties in PD seems to be restricted to sleep onset and early awakenings, rather than sleep maintenance.

Previous reviews highlighted the existing inconsistencies in sleep architecture data. Overall, they report that there is as much evidence showing that the sleep architecture of patients with PD and of healthy controls is different as there is evidence that they are not [3, 40]. The narrative nature of previous reviews limited the conclusions that could be drawn from such data. Given this limit, the present study used meta-analysis to synthesize conflicting data. Results showed no difference between the sleep of patients with PD and the sleep of healthy control participants on the percentages of stage 1 and stage 2 sleep, of slow-wave sleep, and of REM sleep. Also, there were no differences between groups for the number of REM periods, REM sleep density and duration, stages 1 and 2 sleep duration, slow-wave sleep duration, and percentages of stage 3 and of stage 4 sleep. However, the latter results were less reliable due to the small number of studies that were included (two to four studies for each variable). Therefore, confirmation of the results is needed.

In sum, there is reliable evidence that, in comparison to healthy controls, patients with PD demonstrate alterations in objective sleep latency, sleep efficiency, and total sleep time. Also, it is plausible that percentages of REM sleep, of delta sleep, and of

stage 1 and stage 2 sleep do not differ between patients with PD and controls. For other variables, further research is required to confirm existing results.

#### **4.3 Why is sleep altered in PD?**

A recent literature review highlighted the need for studies to go beyond mere descriptions of sleep disturbances and to investigate the role of sleep disturbances in the development and maintenance of PD [3]. Research on the impact of cognitive activity on sleep yielded important insights into the onset of insomnia in PD. Studies of the possible links between repetitive thought (e.g., worry or rumination) and various sleep characteristics in college students indicated that repetitive thoughts impact sleep [49]. Since patients with PD often worry about having panic attacks, it is possible that reported sleep alterations could be associated with repetitive worry about future panic. This hypothesis is consistent with Harvey's [14] cognitive model of insomnia, which proposes that excess negative cognitive activity plays a central role in the maintenance of insomnia.

In patients with PD, the content of repetitive thinking often relates to possible panic attacks and their consequences. Patients develop a fear of anxiety itself, i.e., anxiety sensitivity [18, 50]. Researchers have found that, in patients with PD, anxiety sensitivity is linked to insomnia and, in particular, to longer sleep latency [18]. Authors hypothesized that anxiety sensitivity makes patients monitor their anxiety signs and symptoms and could increase levels of cognitive, emotional, and physiological activation [18]. This state of activation could disrupt the healthy course of sleep [18].

Sleep impairment caused by activation could be the beginning of a cycle where anxiety and insomnia mutually feed each other. Evidence suggests that sleep deprivation induces activation and anxiety in the general population [12] and increases the risk of panic in patients with PD [13]. Therefore, it is hypothesized that a sleep deficit could increase the risk of panic by inducing activation and anxiety. Indeed, since patients with PD fear signs of anxiety, the activation and anxiety caused by the sleep deficit trigger more anxiety that can culminate into panic attacks. These panic attacks could further disrupt sleep because of the activation state they caused and apprehensiveness. Future research should aim at verifying the hypothesis of the vicious cycle between panic and insomnia in order to better understand the cognitive, behavioral, and emotional mechanisms behind the sleep disturbances identified in the present study.

The majority of the results issued from objective data were consistent with cognitive models of insomnia and PD. In contrast, the finding that patients with PD have shorter stage 2 sleep is not intuitive. One possible explanation for this result is the inclusion of patients with NPA. Since NPA usually occur between stage 2 and stage 3 sleep, there may be a relationship between the presence of NPA and decreased stage 2 sleep. An alternative explanation is that shorter stage 2 sleep is an artifact of overall shorter sleep time. Since stage 2 sleep is the longest stage over the course of a night's sleep (i.e., 45–55% of total sleep time; [51]), the changes in sleep continuity observed in our study (longer sleep latency, lower sleep efficiency, and shorter total sleep time) may be attributable to shorter sleep in all three stages. Since stage 2 occupies the majority of the night, it may be the only stage for which the change can be statistically detected. Indeed, the fact that the percentage of stage 2 sleep observed in patients with PD does not differ significantly from that observed in control subjects supports this hypothesis.

Gathering detailed data on the nature of subjective and objective insomnia is a first step toward a greater understanding of the mechanisms (e.g., cognitions, anxiety sensitivity, repetitive thinking, etc.) through which insomnia could

**107**

**4.5 Limitations**

*A Meta-Analysis of Sleep Disturbances in Panic Disorder DOI: http://dx.doi.org/10.5772/intechopen.86306*

**4.4 Nocturnal panic attacks**

of sleep alterations, particularly regarding sleep architecture.

ing still have to be present in order for NPA to occur.

Interpretation of the present findings should take into account the following limitations. Some variables were reported by very few studies. For example, only two studies reported data on PSQI subscales, slow-wave sleep, and stage 1 and stage 2 sleep duration. On the one hand, the limited data increases the risk of biased results, as there is no guarantee that the two included effect sizes are representative of the entire distribution of true effects [32]. On the other hand, a risk inherent to not running the analyses also exists, as it allows for the possibility of vote-counting, which is even more biased [32]. For example, Papadimitriou and Linkowski [40] discussed conflicting results about self-reported sleep efficiency. With only results of individual studies available, the authors could not draw concrete conclusions.

maintain panic symptoms and vice versa. Nevertheless, the present review does not permit further inferences about interactions between sleep and PD. Considering that analyses of PSQI component scales, stages 1 and 2 sleep duration, slow-wave sleep (duration of slow-wave sleep, percentage of stages 3 and 4), and REM sleep (duration, density, number of REM periods) are based on a small number of studies (between two and four), further research is required to confirm the specific nature

This study's secondary aim was to estimate the prevalence rate of NPA. Results indicated that, among patients with PD, 52.1% (95% CI [46.4, 57.7]) experienced at least one lifetime NPA and 27.0% (95% CI [17.9, 38.6]) experienced at least one panic attack in the past month. Furthermore, 40.9% (95% CI [18.1, 68.5]) of patients with PD reported recurrent NPA (at least 1–2/month in the past 6 months) with apprehensiveness about possible future nocturnal attacks. Of the three estimates, the lifetime prevalence obtained here is the most precise and reliable, with a confidence interval ranging from 46.4 to 57.7%. This represents an improvement compared to previously reported estimates: 44–71% in Lee and Douglass [38], 18–69% in Mellman [39], and 33–71% in Papadimitriou and Linkowski [40]. The two other NPA prevalence estimates calculated in our meta-analysis have wider confidence intervals, indicating poorer reliability. The poorer reliability may explain why the prevalence rate is higher using a more restrictive criterion (1/month or/2 months, with apprehensiveness, present since 6 months; 40.9%) than using a less restrictive criterion (one NPA in the past month; 27.0%). These results may be attributable to measurement error rather than to real differences in prevalence. Some authors have proposed that NPA could be linked to abnormalities in breathing regulation and patterns [52]. Previous research has suggested that patients with PD have chronic hyperventilation [53], which would cause CO2 depletion in the blood [54]. The metabolism would then adapt and become more sensitive to small increases of CO2 in the blood [52]. Since non-REM sleep stages imply a reduced breathing rate and, therefore, a rise in CO2, authors propose that they could be sufficient to induce the physiological reactions and physical sensations that are feared by people with PD [54, 55]. Considering that there have been reports linking nocturnal panic to sleep apnea [56], we could hypothesize that sleep apnea could induce even stronger physiological sensations in people with CO2 hypersensitivity. Therefore, they would be more at risk of having NPA. However, the abnormalities in breathing regulation cannot totally explain the presence of NPA. Cognitive components such as beliefs about physiological sensations and interoceptive conditionmaintain panic symptoms and vice versa. Nevertheless, the present review does not permit further inferences about interactions between sleep and PD. Considering that analyses of PSQI component scales, stages 1 and 2 sleep duration, slow-wave sleep (duration of slow-wave sleep, percentage of stages 3 and 4), and REM sleep (duration, density, number of REM periods) are based on a small number of studies (between two and four), further research is required to confirm the specific nature of sleep alterations, particularly regarding sleep architecture.
