Appendix C: Estimating Waste in Frontline Health Care Worker Activities

C. Jane Wallace, RN, PhD
Clinical Outcomes Research Scientist Intermountain Health Care
36 South State Street, Suite 1900
Salt Lake City, UT 84111

Lucy Savitz, PhD, MBA
Senior Health Services Researcher
RTI International
3040 Cornwallis Road
Research Triangle Park, NC 27709

April 2006

Send all correspondence to:
C. Jane Wallace, PhD, RN
2568 East 3020
South Salt Lake City, UT 84109
Voice: (801) 442-3062; (801) 718-7822; Fax: (801) 442-3821
E-mail: jane.wallace@intermountainmail.org, cjwallac@comcast.net

Key Words: quality, waste, inefficiency, lean health care

Abstract

Rationale, Aims, and Objectives

The Agency for Health Care Research and Quality (AHRQ) funded a studyⁱ to examine the factors contributing to waste and inefficiency in health care. Investigation took place at three levels: the community level, the organizational system level, and the frontline level. The latter aspect of the study used structured observation, guided by Toyota Production System (TPS) principles. These observations, completed as part of a separate operational initiative, was designed to estimate the cost of waste in a cross-section of acute hospital worker activities and provide a qualitative description of observed problems.

Method

An observation tool with explicit definitions for categorizing worker activities and rules for estimating the hourly cost of waste were constructed and reliability verified. A single observer shadowed 61 health caregivers for 72 hours in tertiary academic and community hospital settings using structured, nonparticipant observation of worker activities. Data yielded estimates of waste and a qualitative description of problems encountered.

Results

The average cost of waste (i.e., the cost per hour per worker) ranged from USD 7.40, to USD 18.98 across all roles and functions. Overall, workers encountered an average of two problems per hour. Results are provided for specific roles and functions.

Conclusions

Increased attention to operational quality in health care is needed and could potentially decrease costs while increasing patient safety. Implications for poor operational quality and recommendations for action are presented.

Introduction

The Agency for Health Care Research and Quality ¹ (AHRQ) funded a study to examine the factors contributing to health care waste and inefficiency. Investigation took place at three levels: the community level, the organizational system level, and the frontline level. The latter aspect of the study incorporated results from an operational initiative that was guided by Toyota Production System (TPS) principles.^1,2

Projected U.S. health care spending in 2005 is USD 1.9 trillion,³ and macroeconomic estimates of 50% waste across the health care services sector have been reported.⁴ With spending expected to reach $3.6 trillion by 2014,³ we cannot tolerate continued waste and inefficiency in our industry. U.S. health care outcomes are no better, and in some cases worse, than in other countries with less spending.^4-6 Payroll costs—the largest hospital operating expense—increased at an annual rate of 6% per capita in 2004, compared with 0.9% per capita in 1994.⁷

The substantial labor expenses involved in developing skilled frontline health care workers makes understanding and minimizing waste at the sharp end an important research activity. As part of a larger project designed to provide hospitals with financial estimates of waste within their individual settings, this initiative used observational data to estimate the cost of waste and describe problems in a cross-section of health care worker activities. Observations were completed at Intermountain Healthcare (Intermountain) and the University of North Carolina Health Care System (UNC).

Literature Review

Observational studies of health care workers commonly report chaotic workflow and substantial time spent on nonpatient care activities. Whittington observed 20 psychiatric nurses for 178 hours and documented that only 42.7% of their time was spent with patients.⁸ Degerhamer observed nurses before and after a primary nursing care model was introduced to a surgical unit. They reported an increase in direct patient care time from 23% to 61%.⁹ Potter's detailed observation of one registered nurse's (RN's) day-shift work illustrated fragmentation in work activities.¹⁰ Tucker shadowed RNs on hospital nursing units for 296 hours to examine problems at the bedside. The report described "nursing work to be highly fragmented," with an average of 6.5 "operational failures" per 8-hour shift, requiring 9% of nurses' time to resolve and costing approximately $95 per hour per nurse;¹¹ we note that this was the only identified paper that reported the financial impact of waste.

Observation of 14 residents in a Swiss teaching hospital revealed an average of 360 changes in work activities during a 12.5-hour workday; 44% of residents' time was spent performing procedures, while the rest was spent on administrative tasks (21%), traveling/waiting (9%), breaks (8%), personal education (3%), teaching students (1%), and projects (14%).¹² Lurie's nighttime observation of internal medicine house staff revealed frequent work and sleep interruptions, infrequent patient contact, and considerable time spent documenting cases.¹³ Interns' and residents' work activities, studied prior to changes in work-hour rules, indicated rapid movement between activities and less than 35% of time actually spent with patients.^14,15

Random observations of pharmacokinetics residents at a teaching hospital revealed that 36% of their work was related to pharmacokinetics consultations, with the remaining time divided between meetings and teaching.¹⁶ Hollingsworth observed nurses, faculty physicians, and residents in a 36-bed emergency department; overall time spent in direct patient care was 32%, indirect patient care was 47%, and nonpatient care was 21%.¹⁷ While most studies reported detailed activity data for observed workers, they were limited to investigation of physicians, nurses, and pharmacists. Our intention was to study a more varied sample of health care workers.

The two major observation methods used in ethnographic research to understand health care worker activities are work sampling and time-and-motion studies, ^18-35 and there is disagreement regarding the method of choice. We reviewed numerous studies when seeking a suitable method for our research^{8,9,11-15,17,23,24,28,31, 35}; our decision to use structured observation was guided by the need to quantify the time spent in each activity. Data collection in most studies typically involved either continuous or random observation intervals, where activities were timed and detailed notes or a categorical activity list were used.^{8,11-15,24,31,35} After evaluating a number of tools and methods, including the Nurses' Daily Activity Recording System (NURDARS),⁸ Kitson's Therapeutic Nursing Function Matrix,²⁸ the NASA-Task Load Index,²⁴ multidimensional work sampling,¹⁶ studies of primary care internist and pediatrician activities,³⁵ pharmacy functions and pharmacokinetics resident activities,^16,31 hospital house staff (resident and intern) work activities,^12-15,23 cognitive shifts and interruptions in RN work,¹⁰ and unlicensed versus licensed nursing staff activities,³³ we elected to develop a tool.ⁱⁱ

Methods

Observation Tool Development

Worker Activity Measurement

Initially, one observer shadowed two different RNs for 60 and 90 minutes, respectively, using unstructured observations. The observer recorded a description and timed the duration of all activities with a stopwatch. Data were entered into a spreadsheet and a category was assigned to each activity. Nine activity classification categories, consistent with TPS principles, emerged from the data. We included summaries of total observation time, time spent in each activity category, and frequency of location changes and interruptions, along with field notes and a text narrative of each observation. From these data, a structured observation worksheet was developed. 

Table 1. Activity Categories and Definitions

†Patient Safety Resources: Definitions, National Patient Safety Foundation.

During the next 13 observations, four additional activity classification categories emerged, and we documented the mutually exclusive definitions with explicit measurement rules (Table 1). Activity classification categories were reviewed with the observed workers. Three clinicians with research experience also reviewed the definitions. Guided by concepts from the manufacturing literature, we grouped the activity categories into six major classes.

Waste Estimates

We found no quantitative data regarding waste in frontline health care worker activities; consequently, we created rules for a range of estimates. Our estimates were based on the concept of waste (muda) within the context of the TPS.^1,2 In the TPS organization, muda includes defects in products or services, overproduction of unnecessary products or services, unnecessary processing, unnecessary movement of people or goods, waiting, and excess inventories.¹ We assumed that waste in operational activities at the front line of health care is common and generally unrecognized.^11,36 Given our current health care processes, policies, and regulatory environment, some waste may be unavoidable, but until we learn to recognize waste, we cannot seek to reduce or eliminate it.

The following waste estimate rules apply:

1. Assume no waste in time spent on operations.
2. Time spent dealing with defects, errors, locating, waiting, and other categories is 100% waste.
3. Estimate a range of waste (low is 20%, medium is 50%, and high is 80%) for time spent clarifying, processing, stocking, and traveling. A spreadsheet was developed to calculate the waste estimates after each observation was finished (go to Data Collection).

Problem Identification and Coding

Problems documented in the field notes were initially defined as errors, defects, missing supplies, and rework. It was apparent after a few observations that a broader definition was needed. Problems were then defined as an "undesirable gap between an ideal and actual state that hinders a worker's ability to complete his or her tasks, impacts service quality or patient satisfaction."¹¹ Most problems were directly observed; some were reported by the worker as recent (i.e., within the past 24 hours) or recurrent frustrations.

The coding schemes for problems emerged during qualitative analysis of the problem database. Problem categories sometimes overlapped with activity categories. Field notes were sufficiently detailed to determine if the problem disrupted workflow or therapy. Disrupted workflow was defined as interference with the worker's ability to complete the task at hand. Disrupted therapy was defined as interference with a time-sensitive diagnostic evaluation or therapy (e.g., overbooked CT scanner schedule causing delays in emergency patients' evaluations). We used a simplified scale derived from Tucker's research¹¹ to evaluate errors in terms of risk to patients or staff. For each error, the risk was coded as follows: very low—no foreseeable risk, low—error caused patient discomfort, moderate—potential for risk given other conditions being present (example, unclear medication orders), or high—foreseeable potential for harm. 

Data Collection

The principal observer, a doctorally prepared nurse with extensive acute care nursing and research experience, completed all observations during the morning (0600—1200) or afternoon (1201—1800) hours. Workers were asked to conduct their normal routines and were assured that the observer would be unobtrusive. Elapsed time was monitored with a digital stopwatch. Field notes, activity categories, location changes, and interruptions were logged at 1-minute intervals. The data were transferred into a spreadsheet and summary report, and text narrative was reviewed with the worker during a postobservation debriefing. The summary report, through automatic links to the spreadsheet, displayed the frequency of interruptions, location changes, time spent in each activity category, and a range of waste estimates, both in minutes and as a percentage of the total. Problems extracted from the field notes were entered as text into a separate spreadsheet. To verify quantitative data quality, individual data files were double checked when transferred to a summary file.

Sample Size and Setting

A purposive sample of 61 health care workers from Intermountain and UNC Hospitals was selected so we could observe a variety of roles. Both health systems are integrated delivery networks with excellent reputations, although most of the observations (N = 52) were completed at Intermountain. Hospitals included two large (500- to 688-bed) tertiary academic referral centers (one at UNC) and three Intermountain community (200- to 300-bed) hospitals. Prior to our observations, we constructed a list of hospital departments and roles. Resource limitations demanded a focus on major caregiver roles, including physicians, nurses, respiratory therapists, social workers, pharmacists, physical therapists, and various technical workers. Departments included intensive care units (ICUs), medical/surgical units, procedural units (e.g., operating rooms, labor/delivery rooms, cardiovascular labs, endoscopy labs), radiology labs, other laboratories, central processing stations, and emergency departments. Observations were generally scheduled through the department managers; physicians were contacted directly. Participation was voluntary and we obtained verbal consent prior to observation; none of the departments or workers we contacted declined to participate. We also obtained approval of an application for exempt research from the sponsoring (Intermountain) Institutional Review Board.

Data Analysis

Statistica 5.5 (Statsoft, Inc.) was used to summarize descriptive statistics (i.e., frequencies, averages, and 95% confidence intervals). To estimate the cost of waste per hour of observation, multiple public resources (most derived from Bureau of Labor Statistics data up to 2004) and wage data from the Intermountain Human Resources database were used to construct a table (go to Appendix, Table A-1) of hourly base salaries plus 30% fringe benefits for each worker's role (physician salaries did not include benefits).^37-40 We intentionally made conservative assumptions with respect to salaries. The range of waste estimates (low, medium, and high) was the product of the percentage of estimated waste and total hourly salary. Inductive analysis of problems was used to categorize their frequency. Errors and defects were defined a priori; all other problems were defined during data analysis.

Reliability Assessment

To verify the reliability of the method, we compared observations from eight different untrained observers with simultaneous, independent observations from the principal observer. Observed workers included two medical doctors, four RNs, one respiratory therapist, one radiology technologist, one pharmacy technician, and one pharmacist. The observers included two nurses, one pharmacist, one health research analyst, one quality improvement analyst, one radiology technologist, and two students with minimal medical backgrounds. Prior to data collection, the observers were briefly introduced to the observation tool and activity categories. Observations were 30 to 60 minutes in length for a total of 9 hours (1 hour n = 7, 45 minutes n = 2, 30 minutes n = 1). Intraclass correlation for percentage of time spent in operations (0.82, P = .01), waste (0.88, P = .003), frequency of location changes (0.82, P = .007) and frequency of interruptions (0.67, P = .054) indicated good (0.6—0.74) to excellent (greater than 0.74) interrater agreement.⁴¹

Reliability of waste estimates between health care systems was conducted by the principal observer. A comparison of repeated observations in 17 roles (11 at Intermountain, 7 at UNC) and 42 observations (32 at Intermountain, 10 at UNC) yielded comparable overall average estimates (all 95% confidence intervals overlapped) for operations (48% versus 38%), clarifying (19% versus 13%), errors/defects (2% versus 3%), processing (15% versus 24%), motion (15% versus 19%), other (0.1% for both systems), and waste estimates (medium) of 36% versus 27%.

Results

Sixty-one workers were observed for 72 hours (36 morning and 36 afternoon hours). Table 2 summarizes workers' demographic data. Professionals included 8 physicians, 26 nurses, and 8 others. Of the RNs, 5 were ICU/emergency department staff, 10 were non-ICU medical/surgical staff, 5 were operating room/post-anesthesia care unit nurses, 2 were house supervisors, 2 were patient care managers, 1 was a labor/delivery nurse, and 1 was an endoscopy lab nurse. The laboratory workers included two phlebotomists, two medical technologists, and two specimen processors; these workers were grouped together with other technical staff (n = 13). In general, the workers were experienced; only 5 (8%) had less than 1 year and 44 (72%) had more than 3 years of experience in their role. 

Table 2. Worker Demographics and Total Hours Observed

Note: SD = standard deviation; MD = medical doctor; RN = registered nurse; OR = operating room.

Figure 1 summarizes the proportion of total observation time spent in six major activity categories. A table with more detailed activity data is in the Appendix (Table A-2). The average, overall cost of waste (i.e., cost per hour per worker) across all staffing groups was USD 7.40 (low), USD 13.20 (medium), and USD 18.98 (high). Interruptions and location changes occurred at an average (standard deviation, range) rate of 8 (11, 0-80) and 13 (11, 0-58) times per hour, respectively (one technical worker assisting with a cardiac catheterization was uninterrupted during a 30-minute observation).

Figure 1. Activities and Estimated Waste for All Staff

Note: N = 61 observations, 72 hours.

Even though our sample size was limited, subgroup analyses of clarification activities suggested differences between roles. For all workers (see Appendix), the average proportion (with 95% confidence intervals) of the observations spent clarifying was 20% (N = 61, 14%- 25%). For physicians the average was 43% (n = 8, 31%-56%), for supervisory RNs (house supervisors and care managers) the average was 68% (n = 4, 38%-98%), and for technical workers the average was 7% (n = 19, 3%-10%).

The 95% confidence intervals for the hourly cost of waste in the nursing supervisory subgroup (n = 4), under all assumptions, far exceeded those for all other nonphysician staff (n = 41) (e.g., low hourly cost of waste was $9.45—$19.52 versus $4.49—$6.73). The 95% confidence intervals for the medium and high costs of waste in the nursing supervisory subgroup also far exceeded those of all nonphysician, nontechnical staff (i.e., respiratory therapists, pharmacists, social workers, physical therapists [n = 8]), and low costs overlapped only slightly (USD 9.45—USD 19.52 versus USD 2.14—USD 10.07). The house supervisors' time was spent assessing staffing issues, finding information for or directing visitors, traveling, and waiting.

One nursing supervisor confirmed that waiting and traveling were common activities. Care managers spent most of their time on the telephone, in patient care conferences, or looking for information to justify admission or plan for discharge of individual patients. We observed one care manager engaged in rework for a patient discharge that was previously arranged but was delayed because of lacking physician coverage.

Physician and pharmacist observations included waiting, traveling, and clarifying. Fifty-five minutes of one 2.5-hour afternoon observation of a hospitalist was spent waiting for a patient admission. An ICU physician spent 10 minutes traveling in order to spend 2 minutes with an outpatient in his office. Pharmacists spent significant time on the telephone clarifying incomplete, illegible, or potentially erroneous orders.

Documentation (7%) and paperwork (8%), noted to be redundant in 50% (6/12) of the problems classified as rework, were the main processing activities. Although the observation scheme did not quantify all redundancies, a review of the narrative summaries revealed that 26% (16/61) of observations mentioned specific examples of redundant documentation or paperwork. Of these, 22% (14/52) occurred at Intermountain, where advanced information technology was available, and 22% (2/9) occurred at UNC, where manual documentation was predominant. We also noticed frontline workers at Intermountain facilities copying data from computer to paper in order to organize the information to fit their needs. It was common for workers at Intermountain to carry printed reports covered with handwritten notes and data copied from electronic sources. The following are some examples of Intermountain processing redundancies:

• A clerk filed printed copies of electronic nursing shift reports in the paper chart and the medical records staff discarded the files after discharge.
• A pharmacist was observed entering medication use data from the hospital information system into two separate computer applications and on paper.
• A nurse copied preadmission medication lists (available as dictated text in the hospital information system) from a handwritten form to a handwritten kardex and a separate discharge order form.
• A hospitalist spent 29 minutes looking through electronic and paper charts to double check discharge orders and produce a dictated discharge summary. UNC redundancies included the following:
  • A respiratory therapist entered paper documentation into a billing system.
  • Physician documentation was repetitive from day to day and required searching for information recorded by other caregivers.

Almost 10% of all observation periods was spent traveling, and 20% (12/61) of workers spent more than 15% of the observation time traveling. Of the 12 workers, 8 technicians spent an average (standard deviation, range) of 27% (7%, 17%—38%) of time traveling. The house supervisors spent 20% and 32% of time traveling, respectively. Eight workers (13%) spent more than 15% of the observation time waiting. The average for these was 24% (9%, 16—40), and included a radiology technician (40%), hospitalist (33%), operating room circulating nurse (30%), clerk (23%), nursing supervisor (20%), patient care technician (16%), and two RNs (16% and 18%).

A total of 159 problems (12 reported by workers) were documented in 85% (52/61) of the observation periods (Table 3). We calculated an average rate of problem occurrences (two per hour) based on direct observation of 147 problems. Resource limitations precluded a detailed evaluation of the impact of problems on patients or staff. Eighty-six percent (114/133) of problems disrupted workflow and 5% (6/133) disrupted therapy. Twenty-five percent (n=4) of errors were coded as high risk (Table 4). 

Table 3. Problem Frequency

Note: IV = intravenous.

^i. AHRQ Contract No. 290-00-0018, Task 11, L.A. Savitz, Project Director.
^ii. A copy of this tool is available upon request from the corresponding author.