International Journal for Quality in Health Care Advance Access originally published online on October 13, 2006
International Journal for Quality in Health Care 2006 18(6):437-445; doi:10.1093/intqhc/mzl053
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Can computers improve patient care by primary health care workers in India?
1 Johns Hopkins Bloomberg School of Public Health, Department of International Health, 2 Johns Hopkins University School of Medicine, Center for Innovation in Quality Patient Care, Baltimore, MD, USA, and 3 Independent Consultant, Bangalore, Karnataka, India
Objective. The objective was to test whether a decision support technology for non-physicians can increase health care utilization and quality.
Design. Before and after measurements were taken from a systematic random sample of patients and staff at randomly assigned intervention and control facilities.
Setting. The study took place at primary health facilities in rural Tamil Nadu, India
Participants. One thousand two hundred and eighty-six patients and 82 staff were interviewed.
Intervention. A computer-assisted decision support technology was introduced to assist with patient screening.
Main outcome measures. Outcome measures included new patient visits per month, a Global Patient Assessment of Care Index, and health worker attitude variables.
Results. There was a difference of difference of 430 new patient visits per month at the intervention sites (P = 0.005), an increase from baseline of 18% at intervention sites compared with a decline of 5% at control sites. The intervention was associated with significant improvements in a Global Patient Assessment of Care Index (mean difference of difference 7.9, P < 0.001). The largest gains were made in patient communication, technical quality, and general satisfaction with care. The attitudes of public health workers toward the new technology and their jobs did not change.
Conclusions. Decision support technologies have considerable potential to improve coverage and quality of health care for the poor and where there is no doctor, but the unreceptive attitude of public health workers would need to be overcome. Application of these technologies should take advantage of their popularity with patients and the opportunity to work through the private sector.
Keywords: computers, decision support technology, evaluation, health workers, patient satisfaction, poverty, technology, utilization
Address reprint requests to David H. Peters, Room E8132 Department of International Health, Johns Hopkins Bloomberg School of Public Health, 615 N. Wolfe St, Baltimore, MD 21205. E-mail: dpeters{at}jhsph.edu
Accepted for publication September 14, 2006.
In much of the developing world, access to quality health care is limited, and people depend on providers who have limited training or supervision, often from the private sector [1,2]. Even though known low-cost interventions could save millions of lives [3], these interventions are not fully effective because many people bypass the failing public health services where these interventions have been put in place [4]. A number of studies have suggested that improving quality of services can increase utilization in low-income countries [5,6], in some contexts even in the face of higher user fees [7–9]. But public providers often lack the resources and systems to encourage high quality services; while insufficient attention is paid to the preferences of the people, the interventions are intended to benefit. The poor may prefer private and unqualified providers because they may be more accessible, affordable, and responsive to their needs, even if the technical quality of care is questionable [2,4,10]. The result is that many people’s health conditions are inappropriately treated.
The increasing availability of low-cost computer technologies has the potential to improve health care decisions for individuals and populations. Evaluating new technologies can be difficult, often requiring multi-disciplinary approaches [11], and are rarely tested in resource-poor environments. Because the best time to assess a new technology is before it is widely advertised or used [12], we conducted a field trial of a computer decision support technology intended to improve clinical decisions of non-physicians in the developing world using several techniques. We examined whether the introduction of this technology had an impact on utilization levels of services, user perceptions of health services, and health workers’ attitudes towards technology. These are factors that determine whether a new technology will spread [13], though they have been largely neglected in the development of public health programs in low-income countries. Hart’s inverse care law [14] suggests that new services first benefit those who are better off rather than the poor, a phenomenon that has been demonstrated in low-income countries [15], and could limit the effectiveness of a new technology in meeting the basic health needs of the poor. We therefore examined whether the decision support technology would have a differential impact on the poor compared with others.
In developed countries, decision support technologies have consistently been shown to have a positive impact on both the process of health care and patient outcomes [16,17]. The few studies from developing countries have shown that they are safe and potentially useful [18–24]. The impact of decision support technologies on the health system and their impact on patients’ views about health care have yet to be assessed in a controlled field trial in a developing country.
 |
Methods |
|---|
 Top Methods ResultsDiscussionAcknowledgementsReferences |
|---|
The setting
The study was carried out in Salem district (population 3.9 million), which is a rural and impoverished area of Tamil Nadu State. Public sector health services in India have deteriorated considerably in the face of low public spending on health [2,25]. In Salem, public services are provided through 18 hospitals, 71 primary health centers, and 666 subcenters. The primary health centers are intended to provide simple curative and preventive services to 30 000 people, although catchment areas vary considerably. The primary health centers are supposed to be run by a physician, but in many cases, the position has not been filled, or the doctor is not on site. Poor working conditions and low pay make it difficult to retain public health workers at their post, particularly physicians who can supplement or replace their incomes in the private sector. Other staff at the primary health center includes two nurses, up to seven auxiliary nurse midwives, one supervising community health nurse, and other paramedical staff (such as a health education officer). Auxiliary nurse midwives also provide outreach services to a population of about 5000 people and staff the subcenters.
The technology
The decision support technology uses the Early Diagnosis and Prevention System, a computer program developed by a non-profit organization working in India built on commercially available software. The software prompts the operator to take a patient history and uses this to offer screening information for physician referral, laboratory investigations, prevention advice, and simple treatment advice for non-complicated illnesses. The software algorithms were based on multiple international and Indian medical references including World Health Organization guidelines for the Integrated Management of Childhood Illness and tuberculosis treatment guidelines and were then reviewed and approved by a panel of physicians from Kempegowda Institute of Medical Sciences. The hardware included a personal computer, printer, solar electric panel, and uninterrupted power supply. The computer operators, who were high school or recent university graduates with no prior experience in healthcare, were trained on using the software for 1 week. While patients are waiting to be seen by the health worker, the computer operator would assess the patient’s vital statistics and ask a series of questions about the presenting complaint and a review of their physiological systems. Each patient encounter with the computer operator took 10 to 15 minutes.
The main use of the decision support technology is to screen for patients who need a physician and to provide advice for those not needing a physician. At the time of the study, the software offered up to eight possible diagnoses for a patient, but it did use any hierarchical order (e.g. by primary complaint, severity of condition, or probability of diagnosis), and did not use the standardized disease codes of the International Classification of Disease. We also assessed how well the software was able to determine whether or not to refer the patient to a physician. A convenience sample of 1557 consecutive patients was taken from two busy outpatient clinics in areas neighboring the study area to compare assessments made by the Early Diagnosis and Prevention System and an independent panel of physicians. The referral decision was based on whether the patient’s diagnoses or treatment required a practitioner with more training than a primary health center nurse. The need for a prescription drug was a typical indication for a referral. The physicians and computer operators were blinded to patient selection and the results of each other’s assessments.
Sample
The study used a quasi-experimental design to compare measurements at sites where decision support technology was used and at control sites both prior to the use of decision support technology (baseline) and after it had been in use for 6 months (intervention). Of the 71 primary health centers in Salem district, three primary health centers were randomly selected to be in the intervention group, and three were randomly assigned to control site. Surveys of patients and health workers were conducted at baseline, and again after the software had been in operation for six months, coinciding with the same season. For the patient satisfaction assessments, patients were systematically sampled by selecting every second or third patient registering for the clinic (depending on the patient load). Selected patients were not known to the health facility staff, and exit interviews were conducted outside the clinic after they completed their visit to the facility. For patients under 18 years of age, the parent or caretaker was interviewed. The final sample included 1286 patients or caretakers. All the 82 health workers at each of the sites were also interviewed.
Instruments
The scales used to obtain patient perceptions were adapted from the medical outcomes patient satisfaction survey [26]. This final instrument, renamed the Global Patient Assessment of Care Index, includes 50 statements concerning perceptions on the quality of care, with responses (and scores) ranging from strongly agree (2), agree (1), disagree (–1) to strongly disagree (–2). Twenty-four of the items were negatively worded; so the scores were reversed, yielding a scale ranging from –100 (worst possible quality) to 100 (best possible quality), with 0 being neutral. The internal reliability of the index was high (Cronbach’s alpha = 0.91). Subscales were constructed among similar items and weighted arithmetically to yield the same scale (–100 to 100), addressing the following domains: 1—General satisfaction with care; 2—Technical quality of care; 3—Respect for patient; 4—Communication with patient; 5—Financial aspects of care; and 6—Access to care. Assessments of health workers attitudes included questions about the role of computers and information technology in their work and a 17-item questionnaire on motivation factors at work.
Patients were classified into five economic quintiles based on an asset index constructed from questions about household ownership of items [27], using the population-based National Family Health Survey II as a reference [28]. The reference scores were derived from identical questions used with a representative sample of the rural Tamil Nadu population in 1998–99. The concentration index was used as the measure of equality of health services [29]. All questionnaires were pre-tested, translated from English to Tamil, and back-translated into English to ensure consistency.
Intervention and control sites were compared at baseline and post-intervention periods using t-tests for continuous variables, and chi-squared statistics for categorical variables. To test the difference of difference of mean number of patient visits, we used a cross-sectional time-series feasible least-squares regression model with first order serial autocorrelation and non-constant variance in the panels [30]. Multiple linear regression was used to assess independent variables on the Global Patient Assessment of Care Index and the six subscales of patient perceptions, testing the interaction of time (baseline or intervention) and intervention or control status—a difference of differences comparison. Cluster effects due to sampling patients from fixed clusters (i.e. the same clinics) were addressed by using a least-squared dummy variable model, with the clinic variable used as a dummy variable [30,31].
|
Results |
|---|
|
TopMethodsResultsDiscussionAcknowledgementsReferences |
|---|
The comparison between physician assessment and the computer’s recommendation for referral is summarized in Table 1. The software proved to be a conservative screening tool—whereas most of those who needed a referral were referred, many patients who did not need to be seen by a physician were also referred to a physician. Using the physician assessment as a gold standard, the sensitivity of the referral decision was high (80%) though the specificity was quite low (22%), with relatively low positive and negative predictive values (65 and 38% respectively). As a result, the software had a low level of Kappa agreement. The physician’s main diagnosis was always consistent with one of the possible diagnoses suggested by the software (up to eight potential diagnoses were provided), but the computer diagnoses lacked any order of priority, probability, or severity of condition to make further comparisons based on diagnosis.
|
At baseline, the characteristics of the patients from the intervention and control groups were slightly different (Table 2). The patients from the intervention site tended to be nearly 6 years older on average and have 1 less year of education than patients from the control sites. However, similar proportions of patients were women and came from households of similar size. It is noteworthy that the average level of education is quite low, reinforcing the high levels of poverty in the district.
|
Utilization levels
The average monthly attendance of new patients at intervention and control sites shows how utilization increased at intervention sites, while declining modestly at the control sites (Figure 1). The overall average of new patient attendance from the period before the intervention (from January to June 2002) was also compared with the same season during the intervention (from January to June 2003). The results showed the monthly new patient load increased by 18% at intervention sites (from 1808 to 2002 new patients) compared with a 5% decline at control sites (from 2038 to 1932 new patients). Adjusting for first-order serial autocorrelation, the difference of difference results were significantly higher at the intervention sites, accounting for 430 visits per month (P = 0.005).
|
The distribution of outpatient utilization by wealth quintile showed that at baseline both intervention and control sites tended to preferentially serve the poor (concentration indices less than 1) and that this tendency did not change (Table 3). If this distribution were applied to the monthly attendance rates, it suggests that utilization by the poorest two quintiles of the population increased 287 visits per month per facility at the intervention sites.
|
Patient satisfaction
The decision support technology had a positive impact on patient satisfaction. The decision support technology was associated with consistent and statistically significant improvements in the Global Patient Assessment of Care Index, as well as satisfaction in each of the domains of care (Table 4). At baseline, there were no differences in any of the dimensions of quality of care between intervention and control sites. But when implemented, patients at intervention sites had statistically higher levels of perceptions about communications with health providers, technical quality of care, and satisfaction with care. A comparison of difference of difference shows that in all domains except physical access to care and, to a lesser degree, financial aspects of care, there were significant improvements at intervention sites compared with controls from baseline to the follow-up period. Multiple linear regression models that controlled for patient’s age, sex, education, household size, and asset score produced similar results (the adjusted coefficients are shown in the differences of differences estimates in Table 4).
|
Health worker attitudes
The survey of health workers identified no difference in ratings of items about access to information, medical records, or the use of technology (Table 5). The only differences were found for those items that indicated whether a computer was being used in the clinic. The decision support technology did not have a noticeable impact on motivating factors for health workers. When examining their attitudes about tools to do their jobs, working conditions, and personal motivating factors, there were no significant differences between intervention and control sites at baseline or after decision support technology was introduced (data not shown).
|
|
Discussion |
|---|
|
TopMethodsResultsDiscussionAcknowledgementsReferences |
|---|
The results of this study highlight both the challenges and opportunities for computer-assisted decision support technology in low-resource environments. For decision support technologies to be more widely applicable, it will be important that key stakeholders have confidence in the software program. The low Kappa agreement of this technology and physician referral decisions would need to be addressed through software improvements. Wherever the decision support technology is used, it would be worthwhile to have an authoritative reference body regularly review and endorse the diagnostic and therapeutic algorithms to add to its credibility and to facilitate changes to suit local conditions and changing standards of care. If the outputs of a decision support technology do not result in better care, it is counter-productive to use the technology to improve patient satisfaction and increase the number of patient visits.
There are also hardware issues to be considered. The currently high capital costs may require donor assistance in many developing countries. Although desktop computers are less likely to be stolen, they are also less useful for field-based health workers, for whom laptops or personal digital assistants would be more appropriate. The technology for linking data from peripheral and central users would also be needed. In India, data can be sent through phone lines or by mailing diskettes. Safeguards would then be needed to better protect any data that have individual identifiers.
It was clear that most of the regular health workers at the primary health centers were not motivated by the addition of decision support technology (as expected, computer operators were enthusiastic about the program). Supplementary focus group and key informant interviews supported the survey findings and revealed that many health workers saw the decision support technology as additional work and an impingement on their private practice. During focus group discussions with physicians, they acknowledged that the decision support technology could serve ‘as triage,’ but strongly felt that it should not replace physicians and that it created unrealistic expectations among patients about the quality of care. It was also disclosed by some that decision support technology could increase their workload at the public clinic and cut into their private practices. Focus group discussions with allied health professionals tended to view the decision support technology more favorably but often felt they would prefer other technology, such as laboratory equipment. They also pointed out that decision support technologies would be more useful for nurses who worked at subcenters where there are no physicians.
In the short time that decision support technology was tested, it was apparent that patients were very pleased with it. This was manifest in increased utilization of outpatient services and by widespread and consistent improvements in patient satisfaction. One reason may be that the decision support technology gave the patient more contact time with a provider, whereas the average length of time with a physician was exceedingly short, averaging 1 minute, and changing little during the intervention (data not shown). Interviews with community leaders confirmed the popularity of the decision support technologies, who frequently pointed out that the computer was ‘more powerful’ than the physician alone, typically calling the decision support technology ‘more detailed’ and an indication of ‘progress at the primary health center.’ If the decision support technologies are to be disseminated more widely, it would be best to base the expansion strategy on the strength of the positive public response and widely share these findings. Although the results showed little response among health providers, there were key opinion-leaders within the health profession who were also enthusiastic about the technology, another potential asset in expanding the role of decision support technologies.
A number of limitations of this study should be highlighted. One limitation is the small sample size of facilities and health workers. Larger sample sizes would have had a greater power to detect changes in health worker attitudes. Nonetheless, independently conducted key focus group discussions supported the inference that there was little impact on these factors in the study area. There were some differences between the intervention and control sites at the beginning of the intervention, such as having younger patients and higher utilization rates at the control sites, but we were not able to identify any systematic reason for these differences. It is possible that the intervention sites drew patients from some of the control sites, but we note that utilization rates did not decline much in the control sites, and interviewers did not notice this phenomenon in the exit interviews.
Another important limitation of the study was that it was conducted in locations where physicians were present. For ethical and scientific reasons, it first needed to be tested with physicians present, even though non-physicians operated the system. We did not test how well the system would work with outreach health workers who are less comfortable with English and computers. We are thus limited in the generalizations we can make about the efficiency of this software and the use of decision support technologies in environments where a doctor is unable to oversee the care provided. Finally, the study design was not able to pull out components of the intervention that were not due to the technology itself, including the motivation of the computer operators and the length of time spent with the computer operator who were not present at the control sites. Patients could not be blinded to the use of the technology, so it may be that their perceptions of higher quality were due to the presence of the computer itself and not to the care provided subsequent to its use.
Clearly, patients and communities respond well to the technology, though more work is needed to gaining acceptance within the public health community. Given the limitations of using decision support technologies in the public sector, such technologies should also be tested among private providers. They may also be more willing to pay for a technology that enhances their prestige and brings in more patients.
The results of this study support the contention that decision support technologies have considerable potential as a diagnostic aid for non-physicians in low-income countries. Such tools can be used to increasingly standardize care provided by non-professionals, raising the quality of care for millions of the world’s poor who seek care from them. Treatment of pneumonia, diarrhea, malaria, and other common killers could be more effectively applied on a larger scale than under the relatively controlled environments under which they have been tested. A secondary benefit is that more reliable health assessments on populations could be used for public health planning and monitoring. Improving the accuracy of understanding the burden of disease in poor countries would contribute to the ability of public agencies to make evidence-based decisions over scarce health resources.
|
Acknowledgements |
|---|
|
TopMethodsResultsDiscussionAcknowledgementsReferences |
|---|
We are appreciative of the Rockefeller Foundation for providing financial support to this study. We also acknowledge the assistance of the staff from Salem, Kengeri and Bagalur primary health centers who provided care for study patients; the data collection teams from Johns Hopkins Bloomberg School of Public Health, St Johns Medical College, and Kempegowda Institute of Medical Sciences; the George Foundation who provided the technology and personnel for training; and the patients who participated in the study.
|
References |
|---|
|
TopMethodsResultsDiscussionAcknowledgementsReferences |
|---|
- Mills A, Brugha R, Hanson K, McPake B. What can be done about the private health sector in low-income countries? Bull World Health Organ 2002; 80: 325–330.[Web of Science][Medline]
- Peters DH, Yazbeck AS, Sharma RR, Ramana GNV, Pritchett LH, Wagstaff A. Better Health Systems for India’s Poor: Findings, Analysis, and Options. Washington D.C.: The World Bank, 2002.
- Jones G, Steketee RW, Black RE, Bhutta ZA, Morris SS, Bellagio Child Survival Study Group. How many child deaths can we prevent this year? Lancet 2003; 362: 65–71.[CrossRef][Web of Science][Medline]
- Brugha R, Zwi A. Improving the quality of private sector delivery of public health services: challenges and strategies. Health Policy Plan 1998; 13: 107–120.
[Abstract/Free Full Text] - Acharya BL, Cleland J. Maternal and child health services in rural Nepal: does access or quality matter more? Health Policy Plan 2000; 15: 223–229.
[Abstract/Free Full Text] - Akin JS, Hutchinson P. Health-care facility choice and the phenomenon of bypassing. Health Policy Plan 1999; 14: 135–151.
[Abstract/Free Full Text] - Mariko M. Quality of care and the demand for health services in Bamako, Mali: the specific roles of structural, process, and outcome components. Soc Sci Med 2003; 56: 1183–1196.[CrossRef][Web of Science][Medline]
- Chawla M, Ellis RP. The impact of financing and quality changes on health care demand in Niger. Health Policy Plan 2000; 15: 76–84.
[Abstract/Free Full Text] - Haddad S, Fournier P. Quality, cost and utilization of health services in developing countries. A longitudinal study in Zaire. Soc Sci Med 1995; 40: 743–753.[CrossRef][Web of Science][Medline]
- Waters H, Hatt L, Peters DH. Working with the private sector for child health. Health Policy Plan 2003; 18: 127–137.
[Abstract/Free Full Text] - Heathfield H, Pitty D, Hanka R. Evaluating information technology in health care: barriers and challenges. Br Med J 1998; 316: 1959–1961.
[Free Full Text] - Mowatt G, Bower DJ, Brebner JA, Cairns JA, Grant AM, McKee L. When and how to assess fast-changing technologies: A comparative study of medical applications of four generic technologies. Health Technol Assess 1997; 1: 1–149.[Medline]
- Rogers EM. Diffusion of Innovations, fourth edition. New York: The Free Press, 1995.
- Hart JT. The inverse care law. Lancet 1971; 1: 405–412.[CrossRef][Web of Science][Medline]
- Victora GC, Vaughan JP, Barros FC, Silva AC, Tomasi E. Explaining trends in inequalities: evidence from Brazilian child health studies. Lancet 2000; 356: 1093–1098.[CrossRef][Web of Science][Medline]
- Trowbridge R, Weingarten S. Clinical decision support systems. In Shojania KG, Duncan BW, McDonald KM, et al., eds. Making Health Care Safer: A Critical Analysis of Patient Safety Practices. Evidence Report/Technology Assessment No. 43. Rockville (MD): Agency for Healthcare Research and Quality, 2001: http://www.ahcpr.gov/clinic/ptsafety/chap53.htm. Accessed 23 July 2005.
- Hunt DL, Haynes RB, Hanna SE, Smith K. Effects of computer-based clinical decision support systems on physician performance and patient outcomes: a systematic review. JAMA 1998; 280: 1339–1346.
[Abstract/Free Full Text] - Aegerter P, Auvert B, Gilbos V et al. An intelligent computer-assisted instruction system designed for rural health workers in developing countries. Methods Inf Med 1992; 31: 193–203.[Web of Science][Medline]
- Auvert B, Aegerter P, Van Look F et al. A hand-held decision-aid system designed for rural health workers Comput Biomed Res 1986; 19: 80–89.[CrossRef][Web of Science][Medline]
- Chandrasekhar CP, Ghosh J. Information and communication technologies and health in low income countries: the potential and the constraints. Bull World Health Organ 2001; 79: 850–855.[Web of Science][Medline]
- Kahen G, Sayers M. Health-care technology transfer: expert and information systems for developing countries. Methods Inf Med 1997; 36: 69–78.[Web of Science][Medline]
- Kastner JK, Dawson CR, Weiss SM, Kern KB, Kulikowski CA. An expert consultation system for frontline health workers in primary eye care. J Med Syst 1984; 8: 389–397.[CrossRef][Medline]
- Pankhurst RJ. Medical diagnosis in developing countries. Comput Biol Med 1980; 10: 69–82.[CrossRef][Web of Science][Medline]
- Uplekar MW, Antia NH, Dhumale PS. Sympmed I: computer program for primary health care. Br Med J 1988; 297: 841–843.
[Abstract/Free Full Text] - Misra R, Chatterjee R, Rao S. Changing the Indian Health System: Current Issues, Future Directions. New Delhi: Indian Council for Research on International Economic Relations, 2002.
- Marshall GN, Hays RD, Sherbourne CD, Wells KB. The structure of patient satisfaction with outpatient medical care. Psychol Assess 1993; 5: 477–483.
- Filmer D, Pritchett L. Estimating Wealth Effects without Expenditure Data – or Tears: An application to Educational Enrollments in States of India. World Bank Policy Research Working Paper No. 1994. Washington, DC: Development Economics Research Group, The World Bank, 1998.
- Indian Institute for Population Sciences. National Family Health Survey II, Tamil Nadu 1998–99. Bombay: IIPS, 2000.
- Kakwani N, Wagstaff A, van Doorslaer E. Socioeconomic inequalities in health: measurement, computation and statistical inference. J Econom 1997; 77: 87–104.[CrossRef][Web of Science]
- StataCorp. Stata Statistical Software: Release 8. College Station, TX: StataCorp LP, 2003.
- Wooldridge JM. Econometric Analysis of Cross Section and Panel Data. Cambridge, MA: The MIT Press, 2002.
CiteULike 
Connotea 
Del.icio.us What's this?
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||