The prevalence and mechanism of medication errors have been well characterized in the literature. However, no prior studies have investigated the frequency and characteristics of errors in the positive airway pressure (PAP) therapy provisioning process when treating patients with sleep-disordered breathing. Just as medication errors may result in unwanted outcomes, it might be anticipated that errors in providing PAP to patients might lead to suboptimal outcomes. Our study seeks to examine the characteristics and frequency of PAP provisioning errors.
This was a retrospective analysis of a cohort of patients in whom sleep-disordered breathing had been diagnosed and subsequently PAP therapy was prescribed. At a 90-day return visit, the PAP therapy the patient was receiving was compared with the intended therapy. Discrepancies were categorized as either prescribing errors (the prescription did not match the intended modality or settings), or setup errors (the modality or settings did not match the prescription).
The overall PAP provisioning error rate was 8%, with errors most commonly occurring during the set-up process. In univariate analysis, insurance type (P = .003), treatment modality (P = .002), and device brand (P = .05) were associated with error and remained significant in multivariate analysis (model fit P = .002). Compliance, residual AHI, and difference in Epworth Sleepiness Scale were not affected by the presence of error.
PAP provisioning errors are common and might contribute to poor treatment outcomes. Errors might be reduced by standardizing terminology across devices, standardizing prescription forms to improve clarity, and by enhanced quality assurance at durable medical equipment suppliers.
Pena Orbea C, Dupuy-McCauley KL, Morgenthaler TI. Prevalence and sources of errors in positive airway pressure therapy provisioning. J Clin Sleep Med. 2019;15(5):697–704.
Current Knowledge/Study Rationale: The process of providing PAP therapy to patients with sleep- disordered breathing provides opportunities for error during prescribing and during device setup by durable medical equipment providers. Currently, no studies have investigated the frequency and characteristics of errors that occur during the provisioning process.
Study Impact: The overall PAP provisioning error rate was 8%, with errors most commonly occurring during setup. These data suggest a need for enhanced quality control in order to reduce errors and enhance patient safety and possibly treatment effectiveness.
Positive airway pressure (PAP) is considered a preferred therapy for obstructive sleep apnea (OSA), with substantial evidence that it can significantly improve health-related quality of life and reduce health care costs.1–4 It has been estimated that OSA remains undiagnosed in 70% to 95% of patients, and that there could be an annual health care savings of $100.1 billion if diagnosis and treatment of OSA were carried out in those patients.5–8 Annually, $12.4 billion is spent on the diagnosis and treatment of OSA, with $6.2 billion (50%) attributed to PAP therapy and oral appliances.9,10 To maximize heath and economic benefit and reduce the risk of harm, it is important to avoid errors and maximize proper delivery of continuous positive airway pressure (CPAP) therapy. There has been much focus on eliminating wastes and risk of harm by reducing errors associated with medication-based therapy.11–13 In contrast, there has been little evaluation of wastes or harm associated with medical errors involving PAP therapy or other durable medical equipment (DME).
The prevalence and mechanisms of medication errors have been well characterized.14–16 Medication errors can occur at each point in the process, between conceiving a prescription and delivery of the drug to the patient (Figure 1). Prescribing errors occur when the incorrect drug is ordered for a patient. Such errors can include wrong drug for indication; wrong dose, quantity, or route; or prescribing a contraindicated drug; and can occur as a result of illegibility of an otherwise correct prescription.14 Prior to computerized physician order entry (CPOE), prescribing errors were common in outpatient practice, occurring in up to 28% of handwritten prescriptions.17 Dispensing errors occur from the time of receipt of the prescription in the pharmacy to the supply of a dispensed medicine to the patient, and these errors have been estimated to occur in 2% to 5% of dispensed drugs.18–20 Finally, administration errors occur when there is a discrepancy between the drug received by the patient and the drug dispensed, and usually happen in an inpatient setting. A national emphasis on reducing adverse drug events has brought about near-universal CPOE, and more recently e-prescribing has reduced the frequency of dispensing errors. There has been a considerable amount of attention placed on this problem, and after electronic prescribing was implemented, medication errors have declined to a rate of 16 of 100 prescriptions from 38.4 of 100.21
Medication and PAP provisioning errors.
Medication errors have been categorized into prescribing errors, dispensing errors, and administration errors. For the purpose of this study, we adopted a similar scheme for positive airway pressure (PAP) provisioning errors. Although administration errors do occur in clinical practice (eg, a patient uses a relatives PAP device to avoid costs associated with PAP therapy), we did not evaluate the frequency of this error type.
Medication and PAP provisioning errors.
Just as there are medication errors, we considered that provisioning errors related to PAP therapy in sleep medicine practice may not be uncommon. However, no studies have investigated the frequency and characteristics of errors in the PAP therapy prescribing process as it pertains to patients with sleep-disordered breathing (SDB). This study aims to evaluate how often provisioning errors occur in sleep medicine practice, identify the most common types of error, and to learn where in the prescribing process errors most often occur. We will also explore which predictors might be associated with PAP prescription errors, as well as whether PAP prescription errors affect patient compliance, residual apnea-hypopnea index (AHI), or improvement of Epworth Sleepiness Scale score within 90 days of treatment.
This was a retrospective study done in the Center for Sleep Medicine at Mayo Clinic, Rochester, Minnesota. The study protocol was approved by the Institutional Review Board.
All patients were initially referred for a sleep medicine consultation in our sleep center and were evaluated prior to testing by a sleep medicine specialist, staff nurse practitioner, or physician assistant. Diagnoses and treatment plans were provided by our sleep medicine staff after review of testing. We randomly selected electronic medical records from the database of all patients at least 18 years of age who underwent polysomnographic testing between January 2017 and October 2017. We next reviewed the records to find patients who received a diagnosis of OSA, central sleep apnea (CSA), treatment emergent central sleep apnea (TECSA), or obesity hypoventilation syndrome (OHS) based on the International Classification of Sleep Disorders, Third Edition and polysomnography performed in the Mayo Clinic Center for Sleep Medicine. From among these records, we selected charts that met the selection criteria until we met our target accrual of 311 patients. Selection criteria included the following: the patient was initiated on treatment with CPAP, auto-adjusting positive airway pressure (APAP), bilevel positive airway pressure (BPAP) in the S or ST mode, auto-bilevel positive airway pressure (auto-BPAP), or adaptive servoventilation (ASV); the patient had a follow-up visit in our center within 90 days of prescribing a PAP device; and had a PAP download report with device settings and compliance data available for review (Figure 2). The cohort of 311 patients was sized according to our power calculation (see next paragraphs).
Data collection flow chart.
The random electronic search for polysomnograms during the study dates yielded studies for review. Clinical records were reviewed for all inclusion and exclusion factors until we had our target of 311 patient records for final analysis. This required review of 1,700 charts with rejection of 1,389. Of the 311, 298 had clinical records where the treatment modality and settings were complete and explicit. In the remaining 13 cases, the diagnosis was clear but the intended settings were not complete in the clinical notes; in those cases, the initial prescription was used as the “gold standard.”
The sample size was calculated to provide information about error rates by assuming that they would likely mirror pharmacological provisioning errors. Knowing that in the PAP provisioning process we use CPOE and a printed DME prescription, we anticipated 16 errors of one sort or another per 100 attempts at provision for CPAP and APAP devices.21 Because the remaining devices are less commonly prescribed, we hypothesized an associated error rate up to twice as high. We hoped to obtain some estimate of error frequency for all PAP modalities. Estimating the proportion of CPAP and APAP devices at 80% of all prescriptions, a sample size of 311 might yield 40 errors in CPAP/APAP and 20 errors in BPAP/autoBPAP/ASV provisioning, hopefully allowing some quantifiable information about the type and frequency of errors. We did not power the study to evaluate the effect of errors on outcomes, because this was a secondary exploration of this study.
Data Collection and Definitions
Demographic parameters collected include age, sex, race, level of education, language, type of health insurance (private or government), and primary SDB diagnosis.
Recorded data in the sleep medicine specialist's clinical note, the PAP prescription saved in our electronic health record, and the PAP therapy download data in ResScan (Version 220.127.116.1129 from ResMed Inc, Poway, California, United States), EncorePro2 (Version 18.104.22.168 from Philips Respironics, Murrysville, Pennsylvania, United States), or F&P InfoSmart (Version 1.1SP2 from Fisher & Paykel Healthcare Limited, Irvine, California, United States) software were obtained. Information collected from the prescriptions included the intended PAP modality (CPAP, APAP, BPAP-S, BPAP-ST, Auto-BPAP, or ASV) and intended settings: CPAP pressures, CPAP minimum, CPAP maximum, inspiratory positive airway pressure including minimum or maximum levels, expiratory positive airway pressure including minimum or maximum levels, pressure support including minimum or maximum pressures, and respiratory rate as appropriate. Information obtained from the PAP therapy download report included manufacturer of the respiratory device, the actual modality and settings being used, total days included in the assessment, average hours of use per night, number of days used, percentage of days the device was used for more than 4 hours, and residual AHI.
The intended modality and settings documented in the clinic note by the sleep medicine provider were used as the gold standard. In a small number of cases where the details were not found in the provider's notes (13/311), the initial prescription information was used as the standard. The modality and settings from the download report were obtained during a follow-up visit that occurred within 90 days after initiating therapy and were compared with the gold standard. When discrepancies were recognized, an error was identified, and it was categorized according to the following scheme:
Prescribing error: the provider writes a prescription with errors. We identified this error by reviewing the note written by the physician and comparing that to the actual prescription provided to the patient. If there was a discrepancy in modality or settings, an error was deemed to have occurred.
Setup error: the device is set to a different modality or setting from that found on the prescription. This error was identified by comparing the modality and settings on the prescription with the information from the PAP device download obtained during the follow-up visit.
Note that it was possible to have both a prescribing and a setup error.
All continuous data distributions were evaluated for normality using the Shapiro-Wilk test. Data are summarized as mean ± standard deviation when normally distributed, or as median and interquartile range (median [Q1, Q3]) when non-normally distributed. To ascertain the relationship of demographic and clinical variables with the presence of errors we first used t tests and Mann-Whitney U tests for continuous variables and Pearson chi-square or Fisher exact test for categorical variables as appropriate. To explore potential predictors of errors we used binary logistic regression modeling. A value of P < .05 was considered statistically significant. Data analysis was performed using JMP Pro13 (13.1, SAS Institute, Cary, North Carolina, United States) and Wizard for Mac (Version 1.9.21, Evan Miller, Chicago, Illinois, United States).
A total of 1,700 patient charts were reviewed to find 311 that met all inclusion criteria. The most common reason for exclusion was lack of 90-day follow-up data. Other common reasons for exclusion included the patient not having a diagnosis of SDB on the polysomnogram or the patient choosing to pursue a treatment other than PAP therapy, such as a mandibular positioning device or positional therapy. The demographic characteristics of the patients are summarized in Table 1. Seventy-four percent were women and 26% were men with a median age of 64 years. Most of the patients had private insurance (62%). The median AHI was 18 events per hour.
The overall PAP provisioning error rate in this cohort was (8% [95% confidence interval 5%, 11%]), with error most commonly occurring during the setup (71% [95% confidence interval 51%, 85%]) process. Three patients had both types of error, prescribing and setup. Of the 24 errors, the most common type of error was wrong setting (15 [62.5%]), followed by wrong modality (9 [37.5%]). BPAP-ST and ASV were the modalities most likely to be associated with error. In BPAPST, provisioning errors occurred in 2 of 7 (28.6%), and in ASV, provisioning errors occurred in 3 of 11 (27.3%). CPAP was the modality in which errors occurred with the lowest frequency, occurring in only 5 of 134 (4%). We also found that error was more commonly associated with government insurance (16 of 118 [13.7%]) as opposed to private insurance (8 of 193 [4.1%]) (P = .004). Factors not significantly associated with the presence of provisioning errors included age, sex, race, language, or educational status (Table 2). The type of error (prescribing error or device setup error) was not significantly associated with age, sex, insurance type, AHI, device modality, or device brand.
Univariate analysis of clinical features and association with prescribing errors.
Univariate analysis of clinical features and association with prescribing errors.
A binary logistic regression analysis was conducted to examine multivariate predictors of error. Predictor variables considered included treatment modality, insurance coverage, device brand, SDB type, sex, age, educational status, race, and language. However, given the total number of provisioning errors (24), we considered that at most 3 predictor variables should be entered into the model in order to prevent “over-fitting.”22 The predictor variables with the highest correlation with provisioning error in univariate analysis were modality (P = .002), insurance (P = .003), and device brand (P = .05) Table 2. There was no interaction between these variables (all P > .25).23 All three predictor variables remained significant in the multivariate analysis (Table 3), and the fit of the whole model was significant (P = .0002). For the device brand, odds of error were not statistically different between Philips Respironics and ResMed products (P = .2207), but Fisher and Paykel devices had a higher likelihood of error in provisioning compared with Philips Respironics (P = .0282) or ResMed (P = .0096) devices. In order to evaluate whether our choice of predictor variables was unnecessarily restrictive, we also checked a model that included all of the aforementioned predictor variables, and then removed predictors that had significant interactions (SDB by race P = .0220, language by race P < .001, language by SDB diagnosis P < .001). Insurance, device modality, and device brand remained the most important predictors (P = .004, .009, .032, respectively).
Binary logistic regression analysis.
Binary logistic regression analysis.
None of the treatment outcomes measured were statistically significant with respect to the presence of error, including the change in Epworth Sleepiness Scale (P = .44), residual AHI (P = .093), % of days more than 4 hours (P = .93), or minutes of daily usage (P = .23).
This is the first study to evaluate the incidence of provisioning errors in PAP therapy. We found an overall PAP provisioning error rate of 8%, and errors were most likely to occur during the setup process. PAP provisioning errors were more common in patients with government insurance, with setup errors being the most frequent (62.5%). There was an increased risk when BPAP-ST, ASV, or BPAP-S was prescribed compared with CPAP or APAP. Additionally, government insurance and the device brand Fisher and Paykel were found to have higher likelihood of error.
Very few data exist in the literature regarding prescribing error as it relates to other DME. One study evaluated the causes and effects of delays in provisioning respiratory DME equipment to pediatric patients being discharged from the hospital.24 They found that fewer than 20% of DME therapies—including oxygen, nebulizers, and BPAP devices—were delivered within 24 hours, and only 30% were delivered within 1 week of discharge. Underlying risks of delayed delivery included having government insurance, communication errors, and “human errors,” which in their study involved not filing the prescriptions appropriately. However, this study did not evaluate whether the delivered devices and modalities matched the prescriber's intention or prescription eventually provided.
The Institute of Medicine reported in 2006 that at least 1.5 million Americans suffered injuries related to drug treatment.25,26 Previous studies have shown that less than 1% of medication errors resulted in an adverse effect.27,28 However when errors occurred the cost could be more than $2,000 per event.28–30 Although we did not look at monetary cost of PAP prescription errors, interestingly we found there was no significant difference in compliance, residual AHI, or change in ESS from the initial visit to the follow-up visit among patients with compared to those without PAP provisioning errors. Nevertheless, these results pertain only to the population that we studied and the outcomes were measured over a relatively short period of time. Therefore, we are reluctant to speculate on long-term effects of incorrect PAP therapy. For example, if patients with overlap syndrome or neuromuscular diseases were to receive the wrong treatment modality, we do not know if this would adversely affect their health or increase health care cost.31,32 In our cohort none of the patients with overlap syndrome had errors in their prescription, and patients with neuromuscular disease were not included in our study.
When electronic prescribing was introduced it was viewed as a promising strategy for improving patient safety. Abramson et al.21 reported a 1.5-fold decrease in prescription errors for eprescribing adopters in the ambulatory setting. In acute care settings CPOE was found to decrease the likelihood of error by 48%.33 However, reports have also come to the forefront stating that electronic prescriptions have contributed to new types of errors in both primary and secondary care.34–36 Although we generate PAP prescriptions via CPOE, errors related to PAP prescriptions still occurred in 8%. This is similar in magnitude to the medication error rate found in an ambulatory practice using CPOE, which was 16%.21 However, medication error rates vary significantly from one study to another due to varying definitions of error and varied settings in which errors take place, and thus far we represent only one report of errors as they pertain to the PAP provisioning process.14 The conclusion we can draw from our data is that although we use CPOE for all of our prescriptions in order to minimize prescribing errors such as illegible handwriting and variability in settings nomenclature, there still appears to be ample opportunity for error to occur. After the prescription is written, it has to be printed and delivered to the DME provider, where the setup of the device will take place. It is during this setup phase where we found that errors most commonly occur, and there is an increased risk when BPAP-S, BPAP-ST, or ASV is prescribed. The causality of errors during this stage is yet unknown, but it could be due to a variety of factors: different level of knowledge, training or experience among DME suppliers, the divergent terminology from the different device brands in the market that create more confusion among suppliers, or lack of a unified system that allows direct communication between DME and sleep providers. Our data suggest that standardization of PAP modality and settings terminology combined with electronic transmittal of PAP prescriptions to DME providers might significantly reduce PAP provisioning errors.
We also found an association between having government insurance and PAP provisioning errors. In 2011, the Centers for Medicare and Medicaid Services (CMS) implemented the Medicare Competitive Bidding Program (CBP) in nine metropolitan areas and in 2017 it became nationwide. The intent of the program was to improve the effectiveness of the Medicare methodology for setting payment amounts for DME, prosthetics, orthotics, and supplies without jeopardizing patient health.37 However, recent studies suggest the program may not be entirely successful.38–41 Puckrein et al.39,40 concluded that the CBP causes significant disruption among Medicare beneficiaries in acquiring the necessary supplies for self-monitoring of blood glucose leading to increased mortality, hospitalizations and associated costs. In 2015 the National Minority Quality Forum published a complete analysis of the CMS's methodology in which investigators concluded that there were many flaws in the data and methodology presented in CMS's report.40,41 The current quality standards for the CBP address billing, equipment type, and personnel qualifications, but do not address accuracy of PAP setup during provision.42 Our data suggest that there is room for improvement in the accuracy of services rendered by DME providers under contract with the CBP. Unfortunately, we were not able to determine whether errors clustered by DME provider. When providers give a patient a prescription for PAP, not only is there no choice regarding DME provider, but there is also not always a record of which providers patients used. The quality of services rendered, both in terms of accuracy and patient experience, would certainly be an area for future study.
Finally, our data showed that there was a statistically higher risk of error when the DME provider set up a Fisher and Paykel device. This may be misleading. In our cohort, only two patients had this device brand, and one of these had an error during the setup process. In our geographical area, this device brand is less commonly used in sleep medicine practices, which could lead to less familiarity during the setup process. This finding should certainly be examined in a larger study before accepting this as representative.
This study has some limitations. A large number of patients were excluded from our cohort, primarily for lack of follow-up visit. As an international referral center, we serve patients from distant regions in addition to our local population and frequently initiate therapies that are followed up elsewhere. There is a possibility that the error rate could have been biased in either direction because of exclusion of this portion of patients. Interestingly, our study sample was nearly 75% female. During the months of January through October 2017, 59.8% of patients who underwent polysomnography and received the diagnosis of OSA in our sleep center were men, and 40.19% were women. We can only speculate that there may be a bias in who returns for follow-up appointments. A prospective study or an online registry would be needed to reduce these kinds of bias and provide a more accurate measure of error rate and outcomes.
In addition, our results reflect one location and care context. It is plausible that error rates could vary considerably depending on local practices, DME expertise, payer practices, and other manifestations of local health care culture. The type of institution, academic compared with nonacademic, may or may not play a role as well. Barker et al. found that medication error rates were not significantly different between Joint Commission accredited and nonaccredited hospitals and skilled nursing home facilities (P = .82), but the error rates were significantly different from one state to another (P = .04).43 We might expect that differences in setup error rates might differ across DME providers. We had hoped to investigate this in our study, but specific DME provider information is not part of our medical records. Consequently, our results should be interpreted in view of these limitations. Nonetheless, this first report provides a magnitude estimate of error rate that should prompt concern and further study.
Finally, we examined a limited spectrum of diagnoses. Error rates and outcomes of PAP provisioning errors in other disorders such as respiratory failure associated neuromuscular disease, chronic alveolar hypoventilation syndromes, or chronic obstructive pulmonary disease may be quite different, and could be a subject for future study.
Errors occur in every complex system that involves human beings, including health care. We have found that PAP provisioning errors are a common occurrence. In addition to careful and mindful follow-up to detect such errors, implementing system changes and practices will be crucial to improve patient safety and provide high-quality care. Standardizing terminology, improving the clarity of setup software, and automating the connection between prescribing and setup might eliminate many errors. Finally, a clinical quality improvement program that evaluates the accuracy of services and quality of patient experience rendered by DME providers might help reduce errors in PAP provisioning.
All authors have read and approved the final version of this manuscript. The authors report no conflicts of interest.