Correspondence between retrospective and momentary self-reported PTSD symptoms: Evidence for peak and end effects in Veterans

Abstract

Validated retrospective self-report symptom rating scales are recommended for posttraumatic stress disorder (PTSD) screening and treatment. However, such reports may be affected by a respondent’s most intense (“peak”) or most recent (“end”) symptoms. The present study evaluated the correspondence between PTSD symptoms assessed using a standard past-month retrospective rating scale and recorded by ecological momentary assessment (EMA) over the same period and tested hypotheses that retrospective scores would be predicted by peak and end-period momentary symptoms. Male U.S. veterans (N = 35) who served post-9/11 completed the PTSD Symptom Checklist for DSM-5 (PCL-5) at baseline and 1 month later. For 28 days during the intervening period, they received quasi-randomly timed text prompts to complete a modified version of the PCL-5 at that moment. Using multiple regression modeling, controlling for the number of completed EMAs and time (days) since the last EMA, we assessed the predictability of follow-up retrospective PCL-5 scores by (a) the mean of all momentary scores and (b) peak and last-day momentary scores. Retrospective PCL-5 scores were closest to peak scores, d = −0.31, and substantially higher than overall mean, d = 0.99, and last-day momentary scores, d = 0.94. In the regression model, peak symptom experiences and last-day momentary symptoms uniquely predicted follow-up PCL-5 scores over and above significant prediction by overall mean momentary symptom scores. In sum, participants’ self-reported past-month PTSD symptom severity did not simply reflect an average over time. Additional questioning is needed to understand peak and recent symptom periods reflected in these estimates.

The U.S. Department of Veterans Affairs (VA) and Department of Defense (DoD) clinical practice guidelines recommend the use of psychometrically validated, retrospective symptom rating scales when screening for posttraumatic stress disorder (PTSD; VA/DoD 2017). These assessments are also principal tools in measurement-based clinical care (Fortney et al., 2017) and long-term clinical monitoring for PTSD (VA/DoD, 2017), and they have been used extensively in PTSD research. Such scales involve the retrospective rating of symptom frequency, intensity, or impact over a specified period of time. For example, because the Diagnostic and Statistical Manual of Mental Disorders (5th ed.; DSM-5) specifies a symptom duration of at least 1 month in its PTSD diagnostic criteria, both the Posttraumatic Diagnostic Scale for DSM-5 (Foa et al., 2016) and the PTSD Checklist for DSM-5 (PCL-5; Weathers, Litz, et al., 2013) use a 1-month time frame to permit diagnostic assessment. However, studies that have assessed symptoms daily using ecological momentary assessment (EMA) have demonstrated high variability in respondents’ PTSD symptoms within a 1-month period (Black et al., 2016, 2018; Naragon-Gainey et al., 2012; Schuler et al., 2019); thus, it is unclear what a single past-month retrospective score represents. A deeper understanding of how respondents generate retrospective symptom ratings would have implications for the interpretation of commonly used scales. It is possible that ratings represent salient singular experiences rather than a global summary of all experiences; this would suggest a need for additional assessment to capture other dimensions of past-month experiences.

Tversky and Kahneman (1974) famously described individuals’ tendency to use heuristics to simplify the calculations needed to estimate answers to questions such as those posed in the PCL-5 about symptoms over the preceding month. Basing estimates on the most acute or most recent symptoms, termed “peak-and-end” patterns or effects, are examples of such heuristics (Kahneman et al., 1993). Peak-and-end effects have been demonstrated with regard to affective ratings of pleasant and aversive film clips (Fredrickson & Kahneman, 1993), retrospective global reports of pain (Redelmeier & Kahneman, 1996; Schneider et al., 2011; Stone et al., 2000), affect while watching an advertisement, annoyance during exposure to aversive sound, and other assessments (Fredrickson, 2000). However, other retrospective summaries, such as ratings of overall happiness during a vacation lasting an average of 7 days (Kemp et al., 2008), have not demonstrated peak-and-end effects, and such effects have failed to emerge in some examinations of symptoms during shorter time frames, such as past-day fatigue (Schneider et al., 2011) or past-week positive and negative affect (Ben-Zeev et al., 2009).

Across a variety of trauma-exposed and PTSD-diagnosed samples, researchers have examined concordance between momentary self-reported PTSD symptoms and retrospective self-reports of symptoms occurring during the momentary assessment timeframe in different ways (Kleim et al., 2013; Naragon-Gainey et al., 2012; Schuler et al., 2019; Westermeyer et al., 2015). To date, only two studies of which we are aware have specifically considered the effects of peak-and-end experiences on retrospective PTSD symptom ratings. Westermeyer and colleagues (2015) reported on a study of 17 participants diagnosed with PTSD who recorded their most severe individual symptom in each of the DSM-IV Criterion B, C, and D symptom clusters daily for a year. Both the peak daily symptom rating and average symptom rating across days in a given month had similar and strong correlations, though not compared statistically, with a past-month retrospective PCL score. In a study of responders to the September 11, 2001 (9/11), terrorist attacks on the World Trade Center both with and without PTSD, Schuler et al. (2019) did not find support for a unique peak effect on retrospective self-reports. In that study, eight of the 20 PCL-5 items were used to assess momentary PTSD symptoms three times daily for 1 week, with scores compared to retrospective reports of symptoms covering the same 1-week period and using the same eight assessment items. Several summaries of momentary symptoms were compared to the retrospective score, including a single peak momentary score. Bivariate correlations between individual summary measures and the retrospective score were similar and strong, but paired-sample t tests determined the peak score was significantly higher than the retrospective score. The authors concluded that the retrospective ratings more closely reflected periods of more severe PTSD symptoms than a day-to-day aggregate of symptoms occurring during the reporting period. Neither study found evidence for a unique association between most-recent (i.e., end-period) daily symptom reports and retrospective symptom reports using variations of the PCL.

Understanding whether peak and/or end effects impact standard past-month PCL-5 self-report assessments would provide insight into the interpretation and limitations of self-report symptom checklists used to assess the diagnostically relevant past-month timeframe. To our knowledge, the present study was the first to compare momentary scores assessed using all 20 PCL-5 items to retrospective scores using this standard past-month assessment. In addition to evaluating peak-and-end effects by conducting bivariate associations and means comparison, we assessed the unique and combined proportion of variance in the retrospective score that could be accounted for by the peak, end, and mean of all momentary scores. The present study expands the limited peak-and-end PTSD symptom literature by reporting on an examination of these effects in veterans exposed to military-related trauma. Given extensive literature supporting peak-and-end effects on self-reported pain and other subjective experiences, we hypothesized that peak-and-end symptom experiences would significantly predict self-report assessments of past-month PTSD symptoms.

Method

Participants

Data were drawn from an EMA study of PTSD symptoms and high-risk sexual behavior among male veterans. Veterans were recruited between March 2015 and January 2017 from two Veterans Health Administration (VHA) outpatient clinics in Connecticut, in the northeastern United States. Recruitment was carried out via VHA clinicians and staff providing study contact information to veterans, and flyers with study information were also posted in VHA clinics. Participants were required to have served in the United States Armed Forces after 9/11 and to have experienced a military-related traumatic event. In addition, participants had to have recently scored 24 or higher on the PCL-5, with symptoms rated in relation to their military trauma exposure. This minimum threshold was based on the National Center for PTSD’s recommendation to use a cutoff score lower than the suggested diagnostic threshold of 33 when a more sensitive screening measure is desired (National Center for PTSD, 2017). Another study component, not described in the present manuscript, focused on high-risk sexual behavior; participants were required to report at least one high-risk sexual event in the past 28 days, defined as unprotected sex or sex under the influence of drugs or alcohol with a noncohabitating partner, having more than one sexual partner, or trading sex for money or drugs. To facilitate daily EMA completion for 28 days, participants were required to have reliable access to text messaging.

In total, 44 eligible veterans provided written informed consent for all procedures after the initial screening. Eligibility screening was conducted under an institutional review board–approved waiver of written informed consent. Of the eligible individuals, 35 who completed momentary assessments and follow-up PCL-5 assessments were included in the present analysis. Of the nine veterans who were not included, six were withdrawn for failure to complete momentary assessments for 7 consecutive days, an a priori withdrawal criterion; one withdrew from the study due to a work crisis; and two were lost to follow-up after the EMA assessment period. With regard to previously measured characteristics, there were no statistically significant differences between participants included in the present analysis and those excluded due to incomplete data except that those included were less likely to have a history of psychiatric hospitalization (37.14% vs. 77.78%), p = .029. See Table 1 for sample description.

Table 1.

Sample Description

Variable	n	%	M	SD
Age			32.91	6.97
Years of education			14.69	1.37
White/Caucasian	16	45.7
Black/African-American	13	37.1
Hispanic	5	14.3
Asian	1	2.9
Single	16	45.7
Married / remarried	4	11.4
Widowed, separated, or divorced	15	42.9
Months since separation from active duty			50.60	65.19
Served in Afghanistan	23	65.7
Served in Iraq	20	57.1
Served in more than one theater	11	31.4
PCL-5 score at baseline			49.40	12.26
Risky drinking in past 28 days	23	65.7
Any illicit substance use in past 28 days	13	37.1
Lifetime history of psychiatric hospitalization	11	31.4

Procedure

To assess military trauma exposure, veterans completed a modified version of the Life Events Checklist for DSM-5 (LEC-5; Weathers, Blake, et al., 2013). Following the LEC-5, a trained research assistant asked participants to indicate their “worst” traumatic event, including whether the event was military-related, a study inclusion criterion, and whether it involved combat. Questions were followed by completion of the PCL-5 without Criterion A (Weathers, Litz, et al., 2013) to assess PTSD symptoms. Eligible individuals were invited to participate in an in-person informed consent interview followed by a baseline battery of assessments, including those intended to capture data on demographic characteristics, military service history, psychiatric history, and substance use.

Participants received in-person instruction in the study’s EMA methods during their baseline visits. For 28 days following the baseline visit, participants received three daily automated text messages, each randomly timed within a 5-hr interval between 7 a.m. and 10 p.m., yielding a total of 84 messages. Text messages prompted participants to complete a linked survey at that time; surveys could be deferred to another time within the same day.

After the 28-day EMA period, participants completed a repeat administration of the standard (i.e., past-month) PCL-5, without Criterion A questions. They were given a copy of the LEC that they had completed during screening and asked to think about their previously reported worst event when completing the follow-up PCL-5.

Participants were paid $30 (USD) for their participation in the baseline assessment, $25 for the briefer follow-up assessment, $3 for each momentary assessment completed within 15 min of the text prompt, and $2 for each momentary assessment completed outside of the 15-min window but within the same day, with total possible compensation of $307. All study procedures were reviewed and approved by the Institutional Review Boards of the VA Connecticut Healthcare System and Yale University.

Measures

Past-Month PTSD Symptoms

The PCL-5 (Weathers, Litz, et al., 2013) is a 20-item assessment that asks respondents to rate specific PTSD symptoms they have experienced during the past month. Each item describes a symptom from one of the four DSM-5 symptom clusters (i.e., reexperiencing, avoidance, negative alterations in cognitions and mood, and alterations in arousal and reactivity). Respondents are asked to rate how much they were bothered by each symptom over the last month, scoring responses on a scale of 0 (not at all) to 4 (extremely). Total scores can range from 0 to 80, with higher scores indicating higher levels of symptom severity. The suggested score threshold for a probable PTSD diagnosis is 33. In the present sample, Cronbach’s alpha was .84 at screening and .88 at follow-up.

Momentary PTSD Symptoms

Momentary PTSD symptoms were assessed using all 20 items of the PCL-5. Items were framed with the question, “In the past 2 hours, have you experienced any of the following?” Respondents used the same PCL-5 scoring range of 0 (not at all) to 4 (extremely) to rate their responses.

Trauma Exposure

The LEC-5 (Weathers, Blake, et al., 2013) is a self-report measure that is used to assess exposure to 16 potentially traumatic events, with an additional item that allows respondents to indicate “other” traumatic experiences not included in the list. Participants can indicate multiple forms of exposure to any listed event, with response options ranging from “does not apply” to “happened to me.” There is no formal scoring protocol. Two minor revisions made to the LEC-5 for the present study included replacing the “part of my job” response option for each event with “happened before I was 18.” In addition, one item used wording from an older version of the LEC; instead of asking about “sudden accidental death,” participants were asked about the “sudden, unexpected death of someone close to you.”

Data Analysis

Descriptive statistics were used to assess distributions of past-month PCL-5 scores administered at screening and follow-up assessments, within-person means of all momentary PCL-5 scores, calculated as the sum of all PCL-5 scores / total number of completed PCL-5 assessments; last-day mean (i.e., end) momentary PCL-5 scores, calculated as the sum of all PCL-5 scores completed on the last EMA / total number of completed PCL-5 assessments on that day; and participants’ single most severe (i.e., peak) momentary PCL-5 scores.

To ensure that analyses of follow-up PCL-5 scores were not confounded by a systematic effect of repeated assessment during the EMA period, we first estimated the linear slope of momentary PCL-5 scores over repeated observations, allowing slope estimates to vary randomly around the mean. Differences in individual slopes were modeled as a function of (a) baseline PTSD severity, as measured using the PCL-5 and within-person means of all momentary PCL-5 assessments; (b) assessment burden, as measured by the number of EMA assessments completed; and (c) person-level factors hypothesized to affect assessment burden: age, history of traumatic brain injury, and education. To assess the associations between and relative similarity of follow-up PCL-5 scores and momentary PCL-5 data, we calculated Pearson’s bivariate correlations and paired-samples t tests comparing the mean retrospective PCL-5 score and three summaries of momentary PCL-5 data: overall mean, last-day mean (i.e., end), and single most severe score (i.e., peak, informed by Fredrickson & Kahneman, 1993), using a Bonferroni-corrected alpha level of .05/3 = .017 for tests of significance.

Using multiple linear regression, scores from the follow-up PCL-5 assessments were modeled as a function of (a) the within-person mean of all momentary PCL-5 scores; (b) the last-day mean momentary PCL-5 score, rescaled as the deviation from the within-person mean to reduce collinearity; and (c) the single most severe (i.e., peak) momentary PCL-5 score, also rescaled as the deviation from the within-person mean. The model controlled for differences in the number of momentary assessments completed and the number of days elapsed between the last momentary assessment and the follow-up PCL-5 assessment (i.e., a measure of nonoverlap between assessment periods).

For one participant who answered only 18 of 20 items in the follow-up PCL-5, the mean of all completed items was imputed for the two missing items. Only complete momentary PCL-5 data were included in analyses.

Results

On average, participants completed 64.7 (SD = 17.82, range: 16–84, 77.0 %) out of 84 possible momentary surveys. The mean time between the last completed momentary PCL-5 assessment and the follow-up PCL-5 was 4.8 days (SD = 6.09, range: 0–26 days), reflecting mean coverage by the follow-up PCL-5 of 25.2 out of 28 EMA days (range: 4–28 days) given that the PCL-5 assesses symptoms that occur over 1 full month.

The slope of momentary PCL-5 scores was not significantly different from 0, indicating no systematic change in values over repeated measures. Individual slope estimates varied significantly around the mean. Interactions between time and any of the tested predictors did not account for differences in the PCL-5 linear slope.

Participants’ retrospective PCL-5 scores correlated significantly with the mean of all momentary scores, r = .67; last-day mean momentary score, r = .71; and peak momentary score, r = .80. The results of paired-samples t tests showed that only the peak score was not significantly different from the retrospective score (ΔM = 5.09, SD = 11.61). The mean of all momentary scores and the last-day mean of momentary scores both were significantly lower than the retrospective score (ΔM = −15.61, SD = 13.64 and ΔM = −15.18, SD = 13.32, respectively).

In the multiple regression model, tests for collinearity among predictors (i.e., variance inflation factor, tolerance, condition indices) were all well below suggested thresholds. The multiple linear regression model (Table 2) confirmed that retrospective PCL-5 scores were significantly predicted by the mean of all momentary scores, the last-day mean score, and the peak momentary score. The model accounted for 71% of the variance in retrospective PCL-5 scores.

Table 2.

Multiple Linear Regression of Retrospective PCL-5 on Momentary Measures

Variable	B	p
(Constant)	15.79	ns
Number momentary PCL-5 completeda	−0.01	ns
Days since last EMA	0.07	ns
Mean all momentary PCL-5	0.59	<.001
Last-day mean momentary PCL-5b	0.45	.026
Peak momentary PCL-5b	0.51	<.001
R 2	.71
F change in R2	19.66	<.001

Note. PCL-5: Posttraumatic stress disorder checklist for DSM 5; EMA: Ecological momentary assessment; Days since last EMA: Number days between last EMA and follow-up PCL-5.

^aRemoving predictors not statistically significant in the final model (i.e., Number momentary PCLs completed, Days between last EMA entry and past-month PCL-5) does not change model inferences.

^bLast-day mean momentary PCL-5 and Peak momentary PCL-5 were entered as deviations from the mean of all momentary PCL-5.

^aRemoving predictors not statistically significant in the final model (i.e., Number momentary PCLs completed, Days between last EMA entry and past-month PCL-5) does not change model inferences.

Model inferences did not change when analyses were limited to 32/35 veterans who met the PCL-5 diagnostic threshold of 33/80.

Discussion

In the present study, the mean of all momentary PCL-5 scores and, as hypothesized, both the peak momentary PCL-5 score and the last-day mean momentary score significantly predicted past-month PCL-5 scores. This finding provides new support for peak-and-end effect hypotheses in the context of self-reported PTSD symptoms among veterans. Past-month PCL-5 scores were most similar to peak momentary scores, differing by only −5.09 points on a scale of 0 to 80, d = −0.31. Conversely, the mean of all momentary scores and last-day mean score were both significantly lower than the retrospective score by almost 1 full (i.e., pooled) standard deviation, d = −0.99 and d = −0.94, respectively.

Given the limitations of a single global estimate of past-month symptom severity, clinicians might gain important additional insight by asking patients explicitly about the contexts in which they experienced their most severe symptoms, with an understanding that peak experiences may have been discrete episodes. In both clinical work and research, a more complete understanding of symptom experiences could be gained by collecting more frequent measures over shorter time frames that require less retrospection (Stone et al., 2000; Trull & Ebner-Priemer, 2009). Multiple measures can provide information about symptom instability, a dimension that previous research determined to be associated with increased risk-taking, over and above symptom severity (Black et al., 2016). Researchers may find that using a combination of symptom dimensions rather than a single retrospective score improves the prediction of PTSD-related outcomes. However, the insight gained herein should be balanced with the demand for additional assessment.

One limitation of the present study was the delay between participants’ final momentary assessment and the follow-up past-month PCL-5 assessment (M delay = 4.8 days). The effect of last-day measures might have been stronger than observed if past-month assessments had been completed without delay. Similarly, peak-and-end symptoms were those captured by the sampling schedule and might not represent the true peak and end-point symptoms the participants experienced. We acknowledge that the validity of momentary assessments of PTSD symptoms using PCL-5 items has not been fully established. However, their strong correlations with past-month estimates and use in the current study to support a repeatedly demonstrated phenomenon strongly support their construct validity. Finally, the study was conducted with a sample of younger male veterans with military-related trauma exposure who reported engaging in recent risky sexual behavior, many of whom also reported recent risky alcohol use. Alcohol use has been associated with impaired executive function (Day et al., 2015), and we speculate that our study population, whose behaviors have been associated with impulsivity (Black et al., 2015; Dick et al., 2010), might make more use of the simpler peak-effect heuristic than estimates requiring more effort, such as averaging effects over time. It is not clear to what extent the present study results generalize beyond this sample, and conclusions should be considered preliminary.

The salience of peak symptoms in retrospective summary ratings might be moderated by the context in which participants experienced their most severe symptoms. For example, peak symptoms experienced closer in time to the retrospective assessment might more strongly predict retrospective scores than peak symptoms experienced at more distal assessment points. In addition, peak symptoms experienced during an otherwise stable period may be more salient than those experienced during a period of high symptom instability. A retrospective interview could be used to identify contextual factors associated with peak symptoms identified in EMA reports.

In the present study, veterans’ retrospective estimates of their past-month PTSD symptoms closely described their most severe symptom experiences. This finding provides insight into the nature and potential limitations of a single past-month symptom estimate and the salience of outlying acute experiences, over and above an individual’s most recent symptom experiences. The results, if replicated with a representative sample of veterans, suggest the clinical value of discussing veterans’ most severe past-month symptom experiences and the potential value of additional questioning beyond standard assessments if clinicians seek to understand a patient’s more typical symptoms or variability in symptoms.