Identifying Episodes of Back Pain Using Medical
Expenditures Panel Survey Data: Patient
Experience, Use of Services, and Chronicity

This section is compiled by Frank M. Painter, D.C.
Send all comments or additions to:

FROM:   J Manipulative Physiol Ther. 2010 (Oct); 33 (8): 562–575 ~ FULL TEXT


Monica Smith, DC, PhD

Palmer Center for Chiropractic Research,
Palmer College of Chiropractic,
San Jose, CA 95134, USA.

OBJECTIVE:   To explore the correspondence between "Episodes-of-Pain" and "episodes of care" for individuals with back pain.

METHODS:   This study was a secondary analysis of Medical Expenditures Panel Survey (MEPS) 2-year longitudinal data. Individual use and utilization of back pain services were examined across ambulatory settings and providers, and linked to MEPS medical condition data to identify individuals with back pain who do not use or who delay or discontinue utilization of health services for back pain. "Episodes-of-Care" and Episodes-of-Pain were approximated through round-by-round temporal mapping of MEPS back pain utilization events data and medical conditions data.

RESULTS:   Of 10,193 individuals with back pain, approximately one fifth did not actively seek care for their back pain. Utilization of services for back pain (Episodes-of-Care) does not always correspond with an individual's full experience of back pain (Episodes-of-Pain). Upwards of 20% of MEPS respondents who use services for some back pain episodes, reported additional episodes for which they do not use services.

CONCLUSIONS:   These findings suggest that other longitudinal studies based only on data that reflect service use, for example, claims data, may incorrectly infer the nature of back pain and back pain episodes. Many individuals report ongoing back pain that continues beyond their Episodes-of-Care, and many individuals with persistent back pain may use prescription drugs, medical services, and other health services only intermittently.

Keywords:   Medical Expenditures Panel Survey, MEPS, longitudinal analysis, back pain, episodes of care, episodes of pain

From the FULL TEXT Article:


The economic burden of back pain is tremendous, and much interest is directed toward assessing the use of health services for back pain [1] and to better understanding the individual experiences over time of those with chronic back pain. Back pain chronicity has been operationalized in various ways in clinical research. Back pain may be defined as chronic/subacute based on the length of a single continuous episode, e.g. as pain duration greater than 4 weeks [2] or greater than 6 weeks, [3, 4] or as pain that persists for more than 3 months, [5] or defined as chronic for pain lasting at least 6 months in duration. [6] Chronic back pain has also been defined as the number of years since the back problem first occurred, e.g. with chronic defined as 5 or more years, [7] which implicitly recognizes that a chronic back condition may also manifest as repeated “flare-up” acute recurrent back pain episodes over some time frame. [8, 9] As noted in recent reviews, episodes-of-pain or episodes-of-care are also somewhat arbitrarily defined in much of the clinical or health services research on back pain, and there is a growing awareness of the need to conceptualize and analyze care-seeking and care provision across episodes and to consider the long-term consequences of pain, functional status, and other outcomes of care. [10-12]

The Medical Expenditures Panel Survey (MEPS) is a readily accessible source of data on health service use and utilization, and costs of care for self-reported conditions such as back pain. [13] The MEPS public use data files have been analyzed to estimate the health care costs and treated prevalence of back pain, [14-16] however such studies largely have reported only cross-sectional point-in-time annualized estimates, or trends over time, based on analyses of data from a series of MEPS one-year data files. Such time series cross-sectional reports can offer but a limited insight into this problem area, e.g. by comparing the overall volume and costs of ambulatory (outpatient and office-based) service use to inpatient service use for back pain care. The MEPS longitudinal data files have added potential for rendering a focused insight and a more comprehensive understanding of the range and variation of individual experiences with chronic back pain over time. For instance, individuals who utilize health services for a single limited back pain “episode of care” may be distinguished from those with longer ongoing episodes or from those who experience episodic recurrence of back pain. Given also that non-use of services does not necessarily equate to an individual’s being “free of pain”, MEPS data further affords the opportunity to better understand long term utilization behaviors for individuals with chronic back pain, e.g. by exploring the correspondence between “episodes of pain” and “episodes of care” for those individuals.

The purpose of this paper is to describe the as-yet underexplored potential of MEPS data for more detailed and useful examination and understanding of health service use and utilization over time by adult individuals with chronic back pain. Three dimensions of the added utility of MEPS data will be described: Use of the MEPS longitudinal panel structure to examine individual use and patterns of utilization over 2-years; linking across MEPS event file types to examine individual use and utilization across settings and providers; and linking event utilization data with medical condition data in longitudinal MEPS analyses to identify individuals with back pain who do not use, or who delay or discontinue utilization of health services for back pain.


Figure 1

As shown in Figure 1, the Medical Expenditures Panel Survey (MEPS) consists of two distinct but related study designs for data collection: a year-to-year cross-sectional survey design, and a 2-year longitudinal panel design. Estimates of health service use for back pain based on the MEPS cross-sectional annual survey, using full-year consolidated data files, may differ somewhat from those generated by analyses of the slightly more complex data file structures of the MEPS 2-year longitudinal panel survey. I report here specifically my examination based on data from the MEPS longitudinal panels. The MEPS panel survey methodology uses an overlapping panel design, and each MEPS panel survey respondent is interviewed five times over 30 months and asked to recall their experiences during periods ranging from 4 to 6 months. This recall time period was established for the MEPS on the basis of research17 which indicated there is limited recall bias for periods of up to 6 months.

      All 5 Rounds

Table 1

Each MEPS panel uses a 2-year longitudinal study design with participants re-surveyed during each of 5 separate rounds throughout the 2-years. Approximately 94% of MEPS respondents were in-scope and had data for all 5 rounds of the panel (see Appendix Note #1). Table 1 reports the total sample sizes of all MEPS respondents for each of the MEPS Panels 5-through-10, before and after applying the “All 5 Rounds” selection criteria used in this study. Of the 87,302 MEPS participants who were in-scope with data for all 5 rounds, approximately 70% were adults aged 18 or older at the start of the 2-year MEPS longitudinal panel timeframe.

      Back Pain

MEPS interviewers recorded verbatim the respondents’ description of their health care conditions. Trained MEPS coders then re-assigned the verbatim condition narratives to fully specified ICD diagnostic codes which were then truncated to 3-digits to protect confidentiality. In addition to 3-digit ICD condition codes, each condition reported by MEPS respondents is also assigned by MEPS coders to a Clinical Classification Category (CCC), which aggregates health conditions into 259 mutually exclusive and broadly homogenous categories. [18, 19] Chronic conditions coded by CCC categories have been tested and shown to be a useful construct for generating preference-based chronic condition scores for calculating quality-adjusted life years for cost-effectiveness analyses and burden of illness studies. [20] For this study, MEPS respondents with back pain were identified using both the CCC code “205” and the ICD codes “846” and “847”. The CCC code “205” encompasses 66 ICD-9 codes presumed to represent a set of largely chronic back pain conditions (spondylosis, intervertebral disc disorders, other back problems). The ICD-9 back pain codes “846” (sacroiliac sprain/strain) and “847” (other backsprain/strain) are not included in the CCC “205” subset, but instead are categorized within the CCC “232” subset (all sprains and strains, presumably representing acute conditions) Therefore the data capture strategy for this study, was to identify acute, chronic, and recurrent cases in the longitudinal back pain cohort by including all MEPS respondents with any self-reported back pain assigned by MEPS coders to CCC code “205” or coded as ICD code “846” or “847”.

      Ambulatory Back Pain Utilization Events

Table 2

For each of the MEPS longitudinal panels 5-through-10, adult MEPS respondents who were in-scope and had data for all 5 rounds of data collection were identified. Merged data from four separate MEPS event files [21] (MEPS 2009): Office-Based visit events (OB), OutPatient visit events (OP), hospital Emergency Room visit events (ER), and Prescription medications (Rx), were used to identify MEPS respondents who reported a back pain utilization event (visit or prescription) in any of those ambulatory settings. The MEPS Rx events file also included a small number of Rx events for back pain that were only associated with inpatient hospitalizations. The MEPS Appendix Event-Link data files were used to identify and remove those Rx events for back pain identified as being associated with an inpatient hospitalization, which was less than 5% of all Rx back pain events. Reported in Table 2 are numbers of adult respondents with ambulatory-only back pain utilization events, for MEPS panels 5-through-10, i.e. the cohort of “Users” of health services for back pain (N=8,244). Each reported utilization event may have multiple ICD and CCC codes associated with that event, since patients may use a given service on a given date for more than one medical condition (e.g. for related or unrelated comorbid conditions). Reported in Table 2, the back pain utilization events tend to fall into mutually exclusive coding categories, with 90% of events identified solely with code CCC “205”, and 9% identified solely as ICD “846” or”847” code. Less than 1% of the back pain utilization events were coded by MEPS coders as both CCC “205” and ICD “846”,”847”.

      Approximating Episodes-of-Care by Mapping Back Pain Utilization Events across MEPS Rounds

Table 3A

Table 3B

Table 3C

Each MEPS back pain utilization event (OB, OP, ER, and Rx) is identified as having occurred during one of the 5 MEPS data collection rounds. Since prescription medicine refills are not well coded in MEPS, patterns of service utilization for back pain cannot be reliably examined using specific “event dates” (see Appendix Note #2). However, all MEPS utilization events (Rx, OP, OB, ER) can be temporally ordered as occurring in a specific MEPS data collection round, and each round for each MEPS respondent can therefore be classified as either “Back Pain active” (i.e. the round has a back pain utilization event occurring during that round) or “Back Pain inactive (i.e. the round does not have any reported back pain utilization events associated with it). For the cohort of “Users” of back pain services (N = 8,244), Table 3a reports an example of this Round-by-Round mapping of “back pain active” and “back pain inactive” periods across all 5 rounds of MEPS Panels 5-through-10 that may generally approximate back pain episodes-of-care patterns. In Table 3b these same Panels 5-through-10 round-by-round patterns of back pain utilization are collapsed into six general categories of similar pattern types that may represent: those individuals with a single limited back pain episode-of-care (Categories A1 and A2, single back pain round), individuals with a single, possibly prolonged, back pain episode-of-care (Categories B1 and B2, two contiguous back pain rounds), individuals with long-term continuous back pain utilization (Category C, three or more consecutive back pain rounds), and individuals with recurrent episodic utilization (Category D, multiple back pain utilization rounds interspersed with rounds of no utilization for back pain). Also reported in Table 3b is whether the individuals in each of the categories above had only CCC”205” utilization events, or only ICD “846” or “847” events, or both types of events. Individuals with long-term continuous, or episodic recurrent utilization (Categories C and D) predominantly (90%) had been coded by MEPS as having only CCC”205” back pain events. Up to 20% of individuals categorized as a single limited back pain episode (Categories A1 and A2) had only events coded as ICD “846”,”847”. Reported in Table 3c, most individuals with back pain utilized some combination of ambulatory services (OB,OP,ER,Rx), or office-based services (OB) solely.

      Approximating Episodes-of-Pain by Mapping All Back Pain Experienced across MEPS Rounds

The MEPS Medical Conditions file is another source for identifying MEPS respondents with back pain. Medical conditions are assigned corresponding ICD-9 and CCC codes and may be added to the MEPS condition roster in three ways: if the condition is identified as a reason for a utilization event; or if the condition is reported as the reason for any disability days (days missed from work or school); or if the condition is reported as having “bothered” the person during the reference period, i.e. at sometime during the entire round preceding the interview. As an additional measure of disability, each medical condition is flagged if the condition is associated with a day spent in bed.

Table 4A

Reported in Table 4a are the two separate back pain cohorts identified in the Panels 5-through-10 Medical Conditions file. The first cohort of “Users” (n = 8,244) is composed of adults with back pain identified in both the Medical Conditions file and the Utilization Events files, i.e. those individuals with back pain who report using health services for their back pain sometime during the 2-year longitudinal panel timeframe. The second cohort, “Non-Users” (n = 1,949), are adults with back pain identified in only the Medical Conditions file but who are not in the Utilization Events files, i.e. those individuals with back pain who report that they did not use any health services for their back pain during the timeframe of that 2-year longitudinal MEPS panel. Among “Users”, 40% of the cases reported having a disability day due to back pain (missed work/school or in bed), and 34% of “Non-Users” reported having a back pain disability day.

Table 4B

Certain medical conditions, such as back pain, were designated by MEPS as “priority conditions” due to their prevalence, expense, or relevance to policy. Since back pain is a “priority condition”, the data was collected by MEPS and made available in the Medical Conditions file to allow for each MEPS respondent with a back pain ICD-9 or CCC code to be classified into a Round-by-Round pattern that identifies whether any given round is “active” in terms of whether the back pain bothered the person or was associated with a utilization event or disability day during that round. This Round-by-Round mapping is presented in Table 4a for each of the two back pain cohorts (“Users” and “Non-Users”) that were identified in the MEPS Medical Conditions files. In Table 4b is presented the collapsing of the patterns for the “Users” cohort (n = 8,244) into the six general categories approximating experienced episodes-of-pain pattern types, to compare and contrast to Table 3b which presents approximations of back pain episodes-of-care pattern types for this very same cohort of “Users”.


Table 5

Referencing Table 5, approximately 54% of MEPS Respondents (pooled across Panels 5-through-10) are classified as having a single limited back pain episode-of-care, i.e. all of their ambulatory utilization events (Office-Based, Outpatient, ER, and Rx) occur during a single MEPS round. Also reported in Table 5 is the correspondence between the constructed Episodes-of-Care pattern groups (based on MEPS Utilization Event files) and Episodes-of-Pain pattern groups (based on MEPS Medical Conditions file). For individuals initially categorized as a single limited episode-of-care (A) or as single prolonged episode-of-care (B) based only on information about their utilization, approximately 20–25% of these individuals are reclassified into a longer or recurrent episode-of-pain categorization when additional information is factored in (i.e. the individual’s self-report of “being bothered” by back pain or having disability day due to back pain).

Table 6

An individual’s utilization of services for back pain (episodes of care) does not always correspond with that person’s full reported experience of back pain (episodes of pain), in that upwards of 20% of MEPS respondents report additional MEPS rounds during which they experience back pain for which they do not report using health care services (Table 5). For the cohort of “Users” who reported experiencing additional back pain rounds for which they did not use services, the extent to which their additional “Non-Utilization” rounds of back pain tended to precede or to follow their “Utilization” rounds of back pain is examined by mapping the direction of the changes in Round-by-Round patterns as presented in Table 6. For the majority (over 50%), their additional non-utilization rounds occurred subsequent to their utilization rounds (Pattern 1 in 6). For individuals with recurrent, episodic utilization (category D), most of their additional non-utilization back pain rounds tended to “fillin” between their existing utilization rounds.


Of the 10,193 individuals who reported experiencing back pain in this study, approximately one fifth were not actively seeking care for their back pain during the 2-year MEPS panel timeframe. Other longitudinal studies may be based on data that reflect health service use, e.g., claims data, rather than ongoing reports of pain. My findings suggest that such studies based solely on use of health services may incorrectly infer the nature of back pain episodes. Of those who do use services (“Users”), approximately 80% of individuals exhibit the same round-by round patterning of back pain experience and use. For the other 20% of “Users”, however, their additional periods of pain are more likely to occur after their periods of health service use, rather than before. This is consistent with the notion that seeking treatment of some form may begin shortly after the onset of pain, but pain lingers after the receipt of services has ended. As well, this may be an artifact of how the prescription drug data are captured in MEPS, since periods are identified when the drugs are purchased, not when the drugs are taken. Importantly, these data also suggest that many people who might otherwise be classified as having intermittent recurrent bouts of back problems are actually experiencing persistent pain in a continuous series of time periods, but just are not receiving additional services in each period.

I have reported here examples and important considerations for a proposed methodology to examine health service use and utilization for back pain based on analyses of MEPS 2-year longitudinal survey data. Limitations to such a methodologic approach should also be noted. I described in this report an approach to defining back pain based on specific selection criteria of ICD-9 and CCC condition codes, therefore comparability of my findings to other MEPS studies are limited as such. Similarly, I reported limiting my focus to ambulatory utilization events, which guided my selection of relevant MEPS data files, events, and cases, which may also limit the comparability of my study findings to other MEPS studies with different foci, e.g. studies that have described both ambulatory and inpatient health service use for back pain. However, I note that of the 8,244 adult MEPS Respondents (pooled across Panels 10-through-5) that I identified as having any back pain ambulatory utilization events, only 296 (3.6%) of these also had reported any inpatient hospitalization for back pain during the 2-year panel timeframe.

Another limitation is that MEPS is self-report survey data and may therefore be subject to error or bias due to subject recall or other similar problems associated with self reporting of health conditions, health service use, or related impacts such as lost work time or disability. The rigorous MEPS data collection and interview methodology was explicitly designed to reduce the potential influence of non-recall or recall bias as a source of systematic error, though it is important to keep in mind the data collection and data encoding procedures employed in the MEPS. MEPS data are descriptions of health problems that are provided by the MEPS participants in response to prompts from MEPS field interviewers, and the narratives recorded by MEPS field interviewers are then interpreted and assigned by MEPS coders into specific ICD medical conditions and more general CCC coding categories. For those back pain cases that appear in only the MEPS Medical Conditions file but not the Utilization Events files, i.e. individuals with back pain who are not currently using any health services for their condition, it is unknown to what extent, for instance, a vague “self-diagnosed” report of a recurring “bad back” by a MEPS respondent, would be more likely coded in MEPS as CCC “205”, or coded as CCC “232”(sprain/strain), or perhaps some other code altogether. Similarly, MEPS respondents may, over time, receive varying diagnoses of their back problem from health care providers, which may also potentially impact the reporting, interpretation, and coding of specific medical conditions in MEPS. A robust data capture strategy such as I have demonstrated herein, using CCC “205” and the ICD codes for back and sacroiliac sprain/strain to identify back pain cases, yielded a broad inclusive representation of acute, chronic, and episodic recurrent back pain in MEPS.

The collection and encoding of prescription medication data in MEPS represents another potential source of systematic error, in that individuals may carryover unused pain medicine from an earlier back pain episode, or perhaps from a different medical condition entirely, in order to self-treat their current back pain episode (see Appendix Note#2 and Appendix Note#3). It is unknown to what extent this might occur, or even to what extent it may occur in the other direction (i.e. individuals who may use or refill their back pain medications in order to self-treat other medical conditions for themselves or even for others). Such a behavior could also partly explain my finding that “episodes-of-pain” may not always correlate with “episodes-of-care”, in that individuals may self-medicate for a current episode-of-pain by using a prescription leftover from a prior back pain episode-of-care, by refilling a prescription from a different medical condition, or using an over-the-counter medication to self-treat their back pain.

I have proposed herein a categorical method for generically operationalizing the chronicity of back pain in large-scale population-based studies such as MEPS. Although not strictly calibrated as date-specific “episodes-of-care” or “episodes-of-pain”, the differentiated patterns of individual pain experiences and utilization that I described herein may serve useful for better quantifying and qualifying the costs and other impacts related to the experiences and utilization of individuals with back pain. Further research may advise whether this methodologic approach has additional potential utility for similar generic modeling of chronicity in medical conditions other than back pain, particularly for those chronic conditions that may similarly manifest in a recurrent episodic pattern. [22, 23] As well, further research on demographic subgroups may provide useful insights regarding the experience, utilization, and impacts of back pain or other chronic conditions on vulnerable populations and population subgroups of interest. In this study I focused my examination on the utilization and experiences of adults with back pain (see Appendix Note #5 for MEPS study data on children with back pain).

This population-based study contributes insight into better discerning the treated prevalence of back pain, and documents the existence of distinct patient-specific longitudinal patterning of pain experiences and utilization of health services within and across multiple ambulatory care settings. The episodes-of-care and episodes-of-pain proxies generated through a population-based methodologic approach such as demonstrated in this study, may also serve to inform further development and clarification of the presumptive clinical entities of acute, chronic, and recurrent episodic back pain, as these might be more comprehensively defined and applied within an evolving context of cross-disciplinary and integrative clinical practice, and coordinated health care delivery.

APPENDIX:   General data management considerations for examining
back pain using MEPS longitudinal panel data

#1)   All 5 Rounds

Each MEPS panel uses a 2-year longitudinal study design, with participants re-surveyed during each of 5 separate rounds throughout the 2-years. MEPS Panel 9 and onward includes in the dataset a composite variable “ALL5RNDS” which is explained in the MEPS data documentation as a variable to flag participants who were in-scope and had data for all 5 rounds of the panel. After confirming with MEPS staff that “ALL5RNDS” was a new variable not previously available and that MEPS did not plan to retrofit earlier MEPS datasets (Panels 1–8) to include the composite “ALL5RNDS” variable, I obtained from MEPS analysts the SAS programming code they used to create “ALL5RNDS” for Panel 9, translated SAS to SPSS, and verified my SPSS programming code on MEPS Panel 9 data. Approximately 94% of MEPS respondents in Panels 5-through-10 were in-scope and had data for all 5 rounds of the panel.

#2)   Using “Event Rounds” vs. “Event Dates” in longitudinal MEPS Medical Conditions and
Utilization Events files to map patterns of individual experiences with back pain over time

Medical Conditions file   Certain medical conditions, such as back pain, were designated by MEPS as “priority conditions” due to their prevalence, expense, or relevance to policy. For such “priority conditions” MEPS interviewers collected additional information to identify the date the condition began. However, this query was only asked when the condition was first mentioned, therefore only the first occurrence of back pain conditions can be assigned to a specific date using the Medical Conditions file. Since back pain is a “priority condition”, the data available in the Medical Conditions file does allow for each MEPS respondent with a back pain ICD-9 or CCC code to be classified into a Round-by-Round pattern using the Condition Round (CRND) variables. The CRND1-CRND5 variables in the Medical Conditions file identify every round in which that priority condition is reported as having “bothered” the person, or was associated with a disability day (missed work/school or day spent in bed), or was associated with a utilization event for that condition. Utilization Events files   The event dates for MEPS Office-based (OB), OutPatient (OP), and ER visit events are recorded as 3 separate data fields for the “year”, “month”, and “day” of each event, and most OB, OP, and ER back pain events (over 85%) contain valid data in all three of these date fields (year, month, and day). Approximately 15% of back pain event dates contain valid “year” and “month” data, missing only the “day” field. I recaptured these missing event data by imputing “day=15” for the small number of OB, OP, or ER visit events that were missing only the day field. The MEPS prescription medication events (Rx), however, do not have an actual event date, but rather only have a “first ever” Rx date that indicates the first time a prescription medicine was ever taken by a MEPS respondent, even if the first time ever for that prescription preceded the timeframe for that MEPS longitudinal panel. Although each prescription refill during the panel timeframe is included in the MEPS data as a discrete Rx event, refills do not have any “EventDate” associated with the event. Although varying by panel, approximately 15% of Rx events for back pain have a “first ever Rx event date” that occurs within the panel timeframe, and approximately 5% of Rx events for back pain have a “first ever Rx event date” that precedes the panel timeframe. The remaining 80% of Rx back pain events have no data for “first ever date” and are presumably refill Rx events, or original prescription events missing all data for the “first ever” event date. Therefore, most Rx events do not have a useable “EventDate”, either because the “first ever” original prescription date for the Rx event occurred prior to the start of that panel, or because the Rx event is a refill and therefore does not have an “event date”, or because crucial month and year date fields for a first original prescription Rx event are “missing” and therefore cannot be recaptured by imputing “day=15”. However, each and every reported Rx event that occurs during the timeframe of the panel (original prescriptions for back pain, as well as each subsequent refill Rx event for that same prescription medicine), does have an “EventRound” associated with the Rx event, that identifies the MEPS data collection round during which a MEPS respondent reported purchasing a prescription medication for back pain. This feature of MEPS data collection then allows for every reported Rx event to be assigned to one of the 5 MEPS data-collection rounds.

When using MEPS longitudinal panel data to measure health service use and to map patterns of utilization for back pain, there are two important reasons to also include and consider Rx data. First, as demonstrated in Appendix Table 1 (below), there are a substantial number of MEPS respondents who report only Rx events for their back pain, i.e. they have no office-based, outpatient, or ER visits for back pain. If one does not explicitly identify and include those MEPS respondents who report only Rx events for back pain, then one would underestimate (by about 8%) the overall size of the cohort of MEPS respondents who report using ambulatory health services for back pain during the 2-year longitudinal panel timeframe. Second, the inclusion of Rx data is necessary in order to correctly map 2-year patterns of health services utilization for MEPS respondents with back pain. When Rx data is included, approximately 15% of MEPS respondents will “switch” and be reclassified as having a round-by-round pattern with a greater number of “active” rounds with back pain events, compared to their round pattern if only their non-Rx events were included. Appendix Table 2 (below) demonstrates how the inclusion of Rx events redistributes individuals with back pain across the 6-category approximated episodes-of-care schema, pooling across all of the MEPS panels 10-through-5. When Rx events are included, approximately 10% of “single limited episode” MEPS respondents (categories A1, A2) are reclassified as having longer or recurrent pattern of episodes. Similarly, 10–20% of those with a “single prolonged episode” (two contiguous rounds, categories B1, B2), would be reclassified as having a longer or recurrent episode pattern when their Rx events are included. Not including Rx events would therefore introduce systematic bias into estimates of ambulatory back pain utilization or corollary estimates (e.g. costs) based on analyses of MEPS longitudinal panel survey data.

The patterns of health service utilization events for back pain events may also be mapped out month-by-month based on specific event dates, as laid out in Appendix Table 3 (below). As noted, such patterns of back pain utilization should be interpreted more cautiously, since they are devoid of Rx events and therefore underestimate and misspecify back pain utilization to some extent. Nonetheless, the overall data distributions are somewhat consistent when comparing the round-by-round patterns to the month-by-month patterns, in that the percentage of individuals with a single “active” back pain utilization round is also generally consistent with the percentage of individuals with 3 months or less time lapsed between their first and last back pain visit event. Reported in Appendix Table 3, approximately 56% of MEPS respondents (pooled across Panels 5-through-10) incurred all of their non-Rx back pain visits within 3 months or less.

#3)   Detailed sidenote on other minor data cleaning of back pain events and cases used
in the final dataset for this study

As noted in the manuscript, for this study I used the MEPS Appendix Link file to identify and remove from my working datasets those prescription medicine (Rx) back pain events that were specifically associated with an inpatient back pain utilization event. As well, the MEPS Appendix Link file allowed me to identify and remove from my working dataset a very small number of Rx back pain events that were associated with a dental prescription medication purchase, which could represent either a “mis-code”, or perhaps more likely a MEPS respondent who reported purchasing a refill of a pain medicine that had been prescribed by their dentist for a dental problem, in order to self-treat their back pain. As well, I identified and removed from my final working dataset a very small number of cases (n=65) who had a discrepancy in their records, in that they were identified in the Medical Conditions file as having one or more back pain utilization events (Rx, OB, OP, ER, Inpatient, or Home Health) yet these cases did not have a corresponding back pain event recorded in the Utilization Events files.

#4)   Use caution when pooling across MEPS Panels

Caution 1   I identified in my final back pain datasets pooled across MEPS Panels 5-through-10, a very small number of cases that inadvertently shared the same case record identifying number, i.e. the variable DUPERSID had the exact same value in different panels. I confirmed with MEPS that these cases were in fact separate individuals (and not the same individual sampled twice in separate panels). Excerpted here for the reader are the explanation and recommendation from MEPS: “The variable DUPERSID is the combination of the variables DUID and PID. The definitions of Dwelling Units (DUs) in the MEPS Household Survey are generally consistent with the definitions employed for the National Health Interview Survey (NHIS). The Dwelling Unit ID (DUID) is a five-digit random ID number assigned after the case was sampled for MEPS. A person number (PID) uniquely identifies each person within the DU. We (MEPS Staff) have found a few duplicate DUPERSID codes over the years, but it was from different panels and it does not represent the same individual. The DUPERSID uniquely identifies each person within a given panel/year. For example, the DUPERSID “50321018” you have identified represents the same person from HC61 (2001) and HC69 (2002) in panel 6 but a different person from HC97 (2005) in panel 9. It is recommended that you use a combination of panel/year and dupersid to uniquely identify a person when pooling multiple years/panels of data.”

Caution 2   A separate issue worth mentioning here as a caution to analysts pooling across MEPS Panels, is that certain variables may be inconsistently coded across panels. For instance, the MEPS race variable “RACEX” was recoded between years 2001 and 2002. For years 2001 and prior, the value of “1” for the RACEX variable was defined as “white”. For years 2002 and after, the RACEX variable was recoded and redefined so that the value “5” was defined as “white – no other race recorded”.

Caution 3   Analysts who use the STATA statistical software package should keep in mind that the DUPERSID variable in MEPS is 8-digits long, that STATA by default will store 8-digit or longer numeric variables as data type “float”, and that STATA data type float loses precision above 7 digits. That is, an 8-digit DUPERSID variable in MEPS may be “rounded” by STATA to 7-digits, which will create non-unique identifiers for MEPS respondents. The two fixes for this potential problem, setting the DUPERSID data type as “double” or as data type “string”, are described at the website:

#5)   Children with Back Pain

Pooled across MEPS Panels 5-through-10, of the sampled MEPS population who were in scope with data in all 5 rounds (N=87,302), 30% were children aged 17 or younger (n=26,076). Of my total pooled back pain cohort (N=10,753), 95% were adults and only 5% (n=560) were children younger than 18 at start of panel timeframe. Of the 560 children with back pain, 186 (33%) reported having a disability day due to back pain (missed work/school or in bed), 440 (79%) utilized ambulatory health services (Office-Based, OutPatient, ER, Rx) for their back pain during the 2-year MEPS panel timeframe, and only 5 reported any inpatient hospitalization for back pain.

Refer to the Full Text for the Appendix Tables 1-3


  1. Dagenais S, Caro J, Haldeman S.
    A Systematic Review of Low Back Pain Cost of Illness Studies
    in the United States and Internationally

    Spine J 2008 (Jan); 8 (1): 8–20

  2. Chou R, Huffman LH; American Pain Society.
    Nonpharmacologic Therapies for Acute and Chronic Low Back Pain:
    A Review of the Evidence for an American Pain Society/
    American College of Physicians Clinical Practice Guideline

    Annals of Internal Medicine 2007 (Oct 2); 147 (7): 492–504

  3. Fink H (PI):
    Education/Exercise and Chiropractic for Chronic Back Pain. identifier: NCT00561652. processed this record on October 05, 2009.

  4. Smeets RJ, Maher CG, Nicholas MK, Refshauge KM, Herbert RD.
    Do psychological characteristics predict response to exercise and advice
    for subacute low back pain?
    Arthritis Care & Research. 2009 Sept 15;61(9):1202–1209

  5. National Institutes of Health (NIH),
    National Institute for Neurological Disorders and Stroke (NINDS)
    NIH-NINDS Low Back Pain Fact Sheet at web address:
    NIH-NINDS Back Pain Information Page at web page:

  6. Caswell A, West J.
    An investigation into the factors affecting patient selection of chronic low back
    management methods, with particular reference to non-utilization of the
    complimentary therapies, in the United Kingdom.
    Journal of Back and Musculoskeletal Rehabilitation. 2002;16:121–33

  7. Webb R, Brammah T, Lunt M, Urwin M, Allison T, Symmons D.
    Prevalence and predictors of intense, chronic, and disabling neck and back pain
    in the UK general population.
    Spine. 2003 June;28(11):1195–1202

  8. Smith M, Stano M.
    Costs and Recurrences of Chiropractic and Medical Episodes of Low-back Care
    Journal of Manipulative and Physiological Therapeutics 1997 (Jan); 20 (1): 5–12

  9. Hestbaek L, Kongsted A, Jensen TS, Leboeuf YdeC.
    The Clinical Aspects Of The Acute Facet Syndrome:
    Results From A Structured Discussion Among European Chiropractors

    Chiropractic & Osteopathy 2009 (Feb 5); 17: 2

  10. Assendelft WJ, Morton SC, Yu EI, Suttorp MJ, Shekelle PG.
    Spinal manipulative therapy for low back pain:
    A meta-analysis of effectiveness relative to other therapies.
    Annals of Internal Medicine. 2003;138(11):871–81

  11. Hestbaek L, Leboeuf-Yde C, Manniche C.
    Low Back Pain: What Is The Long-term Course?
    A Review of Studies of General Patient Populations

    European Spine Journal 2003 (Apr); 12 (2): 149–165

  12. Koes BW, van Tulder MW, Ostelo R, Burton AK, Waddell G.
    Clinical Guidelines for the Management of Low Back Pain in Primary Care:
    An International Comparison

    SPINE (Phila Pa 1976) 2001 (Nov 15); 26 (22): 2504–2513

  13. Performance Budget Submission for Congressional Justification,
    Medical Expenditure Panel Survey (MEPS)
    Authorizing Legislation: Federal funds pursuant to Title IX and Section 927(c)
    of the Public Health Service Act (PHSA)

  14. Martin BI, Deyo RA, Mirza SK, Turner JA, Comstock BA, Hollingworth W, et al.
    Expenditures and Health Status Among Adults With Back and Neck Problems
    JAMA 2008 (Feb 13); 299 (6): 656–664

  15. Luo X, Pietrobon R, Sun SX, Liu GG, Hey L.
    Estimates and Patterns of Direct Health Care Expenditures Among Individuals
    With Back Pain in the United States

    Spine (Phila Pa 1976) 2004 (Jan 1); 29 (1): 79–86

  16. Feuerstein M, Marcus SC, Huang GD.
    National trends in nonoperative care for nonspecific back pain.
    The Spine Journal. 2004;4(1):56–63

  17. Cohen SB, Burt VL.
    Data collection frequency effect in the National Medical Expenditure Panel Survey.
    Journal of Economic and Social Measurement. 1985;13(2):125–151

  18. Elixhauser A, Steiner CA, Whittington CA, McCarthy E.
    Clinical classifications for health policy research:
    Hospital inpatient statistics, 1995.
    Healthcare cost and utilization project, HCUP-3 research note.
    Rockville, MD: Agency for Healthcare Research and Quality; 1998.
    (AHCPR Pub. No. 98–0049)

  19. Agency for Healthcare Research and Quality (AHRQ),
    Medical Expenditures Panel Survey (MEPS) Access page to Online Workbook:

  20. Sullivan PW, Lawrence WF, Jr, Ghushchyan V.
    A national catalogue of preference-based scores for chronic conditions in the U.S.
    Medical Care. 2005;43:736–49

  21. Agency for Healthcare Research and Quality (AHRQ),
    Medical Expenditures Panel Survey (MEPS) Access page to download data files:

  22. Zhu TY, Tam L-S, Lee VW, Lee KK, Li EK.
    The impact of flare on disease costs of patients with
    Systemic Lupus Erythematosus.
    Arthritis Care & Research. 2009 Sept 15;61(9):1159–1167

  23. Nikpour M, Urowitz MB, Ibanez D, Gladman DD.
    Frequency and determinants of flare and persistently active disease
    in Systemic Lupus Erythematosus.
    Arthritis Care & Research. 2009 Sept 15;61(9):1152–1158


Since 5-22-2016

                       © 1995–2022 ~ The Chiropractic Resource Organization ~ All Rights Reserved