OUP user menu

★ Research Paper ★

Incentives and Barriers That Influence Clinical Computerization in Hong Kong: A Population-based Physician Survey

Gabriel M. Leung MD, MPH, Philip L. H. Yu PhD, Irene O. L. Wong MPhil, Janice M. Johnston PhD, Keith Y. K. Tin BSc
DOI: http://dx.doi.org/10.1197/jamia.M1202 201-212 First published online: 1 March 2003


Objective: Given the slow adoption of medical informatics in Hong Kong and Asia, we sought to understand the contributory barriers and potential incentives associated with information technology implementation.

Design and Measurements: A representative sample of 949 doctors (response rate = 77.0%) was asked through a postal survey to rank a list of nine barriers associated with clinical computerization according to self-perceived importance. They ranked seven incentives or catalysts that may influence computerization. We generated mean rank scores and used multidimensional preference analysis to explore key explanatory dimensions of these variables. A hierarchical cluster analysis was performed to identify homogenous subgroups of respondents. We further determined the relationships between the sets of barriers and incentives/catalysts collectively using canonical correlation.

Results: Time costs, lack of technical support and large capital investments were the biggest barriers to computerization, whereas improved office efficiency and better-quality care were ranked highest as potential incentives to computerize. Cost vs. noncost, physician-related vs. patient-related, and monetary vs. nonmonetary factors were the key dimensions explaining the barrier variables. Similarly, within-practice vs external and “push” vs “pull” factors accounted for the incentive variables. Four clusters were identified for barriers and three for incentives/catalysts. Canonical correlation revealed that respondents who were concerned with the costs of computerization also perceived financial incentives and government regulation to be important incentives/catalysts toward computerization. Those who found the potential interference with communication important also believed that the promise of improved care from computerization to be a significant incentive.

Conclusion: This study provided evidence regarding common barriers associated with clinical computerization. Our findings also identified possible incentive strategies that may be employed to accelerate uptake of computer systems.

Computers and informatics have become important tools in health care delivery. Their applications to the administrative tasks required of a busy health care environment have already demonstrated benefits in terms of the efficient management of patient information.1 For clinically-oriented functions such as electronic medical records and structured knowledge-based clinical decision support, there are suggestions that the positive impact on quality of care and related patient outcomes may be even greater.2

As a result, computers have proliferated in hospitals and clinics in developed Western countries.2 However, in Hong Kong and much of Asia, the adoption of such technology has been disappointingly slow. For example, we previously reported that only about half of local doctors had computerized any part (clinical or administrative functions) of their practice. The situation was most acute for those working solo or in small community-based group practices; 70% had yet to computerize any clinical function.2 These providers are responsible for 85% of all care delivered in the ambulatory setting.3

Previous research58 has postulated a wide spectrum of potential barriers that may impede the adoption of computers in clinical practice. Examples include situational or economic factors, lack of cognitive or physical skills to implement and use technology, medico-legal or liability concerns, and attitudinal or behavioral issues.5 Moreover, we have been unable to locate previous examinations of these factors in the literature as perceived by end-users of computer systems such as physicians.5 This is particularly important for ambulatory solo or group practices, in which the doctors themselves make such entrepreneurial decisions as purchasing and implementing computer systems.

Therefore, to understand the low penetration of information technology among local practices and the potential barriers doctors encounter, we surveyed a representative sample of Hong Kong physicians in 2001. In addition to studying the barriers to computer use for everyday clinical and administrative tasks, we aimed to explore possible incentives or catalysts that would facilitate the adoption of such technology in practice.


Subject Recruitment

This study was a follow-up of an earlier postal survey on the computerization of clinical practice in Hong Kong.2,4 The original survey population (n = 4,850) was randomly selected from all 9,817 physicians on the full and limited registration lists of the Hong Kong Medical Council, the local statutory licensing board, as of 1999. All 949 doctors who responded to that survey were included in the present study. They were representative of the general physician population of Hong Kong, as previously documented.2

The mailing and follow-up strategies were standardized as described previously.2 Briefly, all questionnaires were accompanied by a cover letter explaining the purpose of the study and an assurance that responses would be kept confidential. A prepaid business reply, self-addressed envelope was enclosed for return of the completed questionnaire. First and second reminders, consisting of the questionnaire and a reminder letter, were sent to all nonrespondents who had not replied after 14 and 28 days respectively. As a last resort, trained telephone interviewers contacted those who had not replied after the second mailed reminder. Each subject was contacted by telephone a maximum of seven times during different periods of time (three weekdays, two weekends, two weeknights). If still unsuccessful, the subject was recorded as a nonresponse. All mailings were sent by first-class metered mail. Returned questionnaires with address corrections were mailed to the updated address. We excluded from the analysis mail returned because of unknown or incorrect address and discounted them from the denominator in calculating the response rate. Retirees who indicated on the returned questionnaire or via telephone interview that they were no longer involved in clinical practice were similarly excluded.

As part of a substudy examining the efficiency of cash incentives in eliciting survey response, the 949 potential subjects were randomized under allocation concealment. Half received HKD$20 (USD$1 = HKD$7.8) with the initial mailing, and the other half received the monetary reward on receipt of a completed questionnaire. The questionnaire survey and study protocol were otherwise identical for the two groups. These results are reported separately elsewhere.

Survey Instrument

The questionnaire contained 17 questions divided into four sections. The first section assessed the current extent of computerization of 15 specific clinical and administrative tasks in the main place of work and future intentions to computerize those functions. The second section asked respondents to rank the top five barriers that they perceived to be the most important to the computerization of clinical practice from a list of nine potential factors. No ties were allowed. Similarly, respondents were asked to rank the top five incentives or catalysts from a list of seven factors. The third section dealt with willingness-to-pay questions using contingent valuation techniques (a form of economic evaluation) to elicit how much physicians valued the potential benefits of computerization. The last section covered practice details, such as the size and nature of the practice, working hours, and remuneration patterns.

Questions were formulated and compiled after reviewing other similar reports in the literature,58 current policy directions of the government, and prevailing circumstances of the local health care system.23 In particular, we adopted Johnson's5 conceptual framework for understanding barriers that impede information technology in compiling the list of nine potential barriers. As for the seven incentives and catalysts variables, consultation with local and overseas experts in technology diffusion, medical informatics, organizational management, and health policy analysis informed the drafting and refinement of the list of factors because of the lack of a published, broadly accepted conceptual framework (Table 1). An initial draft of the 17 questions was generated and proposed by the project team, after which the panel of experts gave feedback and confirmed face and content validity from their perspective. All questions were then piloted for comprehensibility and face and content validity on an independent sample of 1,000 physicians (response rate = 7%) randomly selected from the 4,967 physicians not included in last year's initial survey. Minor adjustments were made to the first draft after the consultation and pilot study. Although we have followed a rigorous survey design methodology in compiling the questionnaire, it is difficult to ascertain completely the validity and reliability of the instrument. First, there is no gold standard outcome against which the questions can be benchmarked. Second, the ranking procedure and the complex statistical analysis render such a task virtually impossible.

View this table:
Table 1

Relative Ranking of Barriers and Incentives/Catalysts (n = 731)

Factors (by rank order)Mean Score*SD
1. Time costs associated with computerization including planning, purchasing, training and maintenance (“time costs”)6.162.17
2. Lack of technical support in case of system failure (“lack of technical support”)5.822.20
3. Initial capital (monetary) cost of computerizing a medical practice (“capital costs”)5.702.51
4. Maintenance costs (monetary) including computer, software upgrades and staff training (“maintenance costs”)5.462.11
5. Lack of knowledge and perceived difficulty in learning new technology (“knowledge gap”)4.972.45
6. Interference with the nature and pattern of doctor-patient communication (“interference with communication”)4.852.42
7. Increased potential for breach of confidentiality of patient data (“confidentiality breach”)4.722.38
8. Lack of perceived benefits from computerization of clinical practice (“lack of benefits”)4.082.11
9. Recent amendment of the Intellectual Property Ordinance to prevent copyright piracy (“intellectual property regulations”)3.231.48
1. Improved efficiency with real-time electronic access to laboratory results and hospital inpatient data (“improved efficiency”)5.211.76
2. Improved quality care through better access to medical evidence and information electronically (“improved quality care”)4.691.70
3. Competitive peer pressure in terms of more practices becoming computerized (“competitive pressures”)4.041.80
4. Increased savings from efficiency gains in office operations (e.g., billing and inventory control (“increased savings”)3.901.75
5. Government financial incentives to support the computerization of medical practices (“financial incentives”)3.751.91
6. Increased public or patient demand for a portable electronic medical record and more generally for a computerized practice (“patient demand”)3.491.70
7. Government regulation requiring manda-tory reporting of patient information (“government regulation”)2.911.69

SD = standard deviation.

  • * Refers to the average score for each item, derived from calculating the mean of individual scores in the survey population.

Statistical Analysis

We converted the rank data provided by the respondents into an overall score. For each respondent, the most important barrier identified was given a score of 9, the second most important a score of 8, and so on for the top five barriers cited by the respondent. The remaining four unranked barriers were assigned a score of 2.5 (arithmetic average of the notional scores of 4, 3, 2, and 1). Similarly, the same procedure was applied to the incentives/catalysts ranking responses; the top factor scored 7 and the two unranked variables scored 1.5 each. If respondents ranked fewer than the requested five items, all unranked factors were assigned a score equal to the arithmetic mean of the remaining notional scores. This ranking scheme was devised a priori and is in keeping with generally accepted, conventional techniques in analyzing such data.

We then used multidimensional preference (MDPREF) analysis,910 a graphical method, to examine the relationship between individual subjects and their respective preferences on barrier and incentive items. The data used in MDPREF analysis can be in the form of paired comparisons or rankings of items. In our case the data were expressed as ranks. MDPREF analysis utilizes the technique of singular value decomposition (SVD) and results in a plot in which items are represented by points and subjects as vectors in the same space. The vectors and the points are chosen so that the projections of the items onto any one vector indicate the ranks given the corresponding subject in a low dimensional space as closely as possible. Analogous to principal component factor analysis, a scree plot of the eigenvalues obtained in SVD is used to determine the optimal number of dimensions. Biplots were generated to graphically display the results for the barriers and incentives/catalysts data separately. Therefore, the determination of the key dimensions was entirely data-driven, whereas the actual naming of these dimensions was done by the authors according to the spatial positions or clustering of the vectors representing each individual factor.

Next, we explored potential predictors of respondents' self-perceived importance of the various key dimensions identified from the MDPREF procedure in terms of sociodemographics (age, sex, income, specialist qualifications), practice patterns (work setting and working hours), and other related factors (number of clinical or administrative functions already computerized, competence in computers). Given the requisite linearity assumption in regression analyses and the results from the MDPREF analysis indicating the lack thereof, we instead employed a hierarchical cluster analysis in which no such assumptions are necessary.11 A cluster analysis is a multivariate technique that seeks to identify homogenous subgroups of cases, in terms of the outcome of interest in a given population. That is, it aims to identify a set of groups that both minimize within-group variation and maximize between-group variation. We focused on the variations in the key dimensions identified from the MDPREF analyses for barriers and incentives/catalysts. The average linkage method was employed, and pseudo F and pseudo t2 statistics were used to judge the optimal number of clusters. To identify respondent characteristics that were associated with each key MDPREF dimension from the cluster analysis, we first tested for differences in these characteristics between all the resultant clusters with chi-square and one-way ANOVA. We also compared these characteristics for the two clusters with the most extreme values on a particular key dimension, using chi-square and t-test for categorical and continuous variables, respectively.

We performed a canonical correlation analysis12 to further determine the relationships between the different barriers as a collective set of factors and the corresponding set of incentives/catalysts. A canonical correlation is the correlation of two canonical or latent variables, one representing a set of independent variables, the other a set of dependent variables. The purpose of canonical correlation is to explain the relation of the two sets of variables, not to model the individual variables, by finding a linear combination of each set of variables that yields the highest possible correlation between the composite variable for each of the two sets. Wilks' lambda was used to test the significance of canonical correlation.

To ensure that the prepayment and the postpayment groups had similar baseline profiles, subject characteristics in the two arms were examined using the chi-square statistic. Similarly, bivariate associations between subject characteristics and whether they responded to the survey were analyzed to determine generalizability to the overall doctor population in Hong Kong. Lastly, we examined for differences in the content of the response (i.e., outcome measures) between the pre- and post-payment groups to confirm that the randomization process and payment interventions did not influence the outcomes of interest.

All analyses were carried out using SAS 8.0 and STATA 7.0.

The study received ethics approval from the Research Ethics Committee of the Faculty of Medicine, University of Hong Kong.


Subject Characteristics

Of the 949 questionnaires sent, 731 were returned, achieving an overall response rate of 77.0%. Table 2 shows subject characteristics according to the timing of monetary rewards. Chi-square tests across all baseline factors revealed no significant differences between the prepayment group vs. postpayment group. These results confirmed the homogeneity of respondents across both groups and that our randomization procedure was valid. Furthermore, we found no significant differences for outcome measures in terms of relative rankings or MDPREF dimension scores for barriers or incentives/catalysts between the pre- and postpayment arms (data not shown).

View this table:
Table 2

Subject Characteristics and Bivariate Comparisons

Pre-payment n (%)Post-payment n (%)P value*Respondents n (%)Nonrespondents n (%)P value
   Male317 (81.9)265 (77.9)582 (80.1)167 (77.0)
Years of work experience0.990.08
   0–10100 (25.8)82 (24.1)182 (25.0)58 (27.0)
   11–20141 (36.4)127 (37.4)268 (36.9)67 (31.2)
   21–3089 (23.0)80 (23.5)169 (23.3)47 (21.9)
   >3057 (14.7)51 (15.0)108 (14.9)43 (20.0)
Work setting0.790.01
   Individual135 (37.8)116 (36.8)251 (37.4)98 (47.3)
   Corporate222 (62.2)199 (63.2)421 (62.7)109 (52.7)
Monthly income (HK$)0.63
   <100,000119 (31.8)123 (36.8)242 (34.2)N/A#
   100,000–149,999129 (34.5)105 (31.4)234 (33.1)
   150,000–199,99953 (14.2)47 (14.1)100 (14.1)
   >200,00073 (19.5)59 (17.7)132 (18.6)
Qualified specialist0.230.27
   Yes222 (62.2)182 (57.6)404 (60.0)109 (55.6)

Note: Total number of subjects for each variable may be fewer than 731 and 949, respectively, for payment intervention and response status due to missing values.

  • * This comparison was performed to demonstrate that there were no significant differences between pre- and postpayment groups; hence the randomization procedure was valid and unlikely to have influenced the main results.

  • This comparison was carried out to examine the generalizability of the respondent sample.

  • “Corporate” refers to practice settings in the Hospital Authority, Department of Health, corporate commercial health providers, and the two local universities. All other settings are classified as “individual.”

  • # Data not available.

Table 2 also compares baseline characteristics of respondents and nonrespondents to determine generalizability of the sample to the general physician population of Hong Kong. There were no significant differences in gender (p = 0.32), years of work experience (p = 0.08), and specialist status (p = 0.27), although a higher proportion of respondents worked with corporate employers than did nonrespondents (p = 0.01). We did not have income data from the previous survey; therefore, this variable was not included in these analyses.

Mean Rank Scores and MDPREF Dimensions

Table 1 demonstrates that the three biggest barriers to clinical computerization were time costs, lack of technical support, and large initial capital investments, whereas the lack of perceived benefits and software copyright piracy regulations were ranked the lowest. On the other hand, respondents found improved efficiency and higher quality of care to be the most attractive incentives for computerizing their clinical practice, but patient, public, and regulatory demands for computer systems were ranked as the weakest catalysts.

MDPREF analyses revealed that three key dimensions explained 57.8% of the variance for the barrier variables: (1) cost (time, monetary and personal) vs. noncost factors; (2) physician-related vs. patient-related factors; and (3) monetary vs. nonmonetary cost factors. Figure 1 shows the distribution of the nine barriers in two two-dimensional biplots. Similarly, we derived two key dimensions accounting for the incentives/catalysts (variance explained = 50.7%) from Figure 2: within-practice or internal vs. governmental or external factors and “push” vs. “pull” factors.

Figure 1

MDPREF biplots for barriers.

Figure 2

MDPREF biplot for incentives/ catalysts.

Cluster Analysis

Average linkage cluster analysis for barriers separated the sample into four clusters of respondents (Table 3). Cluster 1 appeared to be most concerned about cost (both monetary and nonmonetary) but relatively indifferent to the other two key dimensions. Cluster 2 focused on nonmonetary costs and, to a certain extent, patient-related factors. Physicians in cluster 3 paid equal attention to patient-related factors and monetary costs, whereas those in cluster 4 cared most about self (physician)-related factors. In terms of respondent characteristics associated with each of the key dimensions, we found that more junior physicians, those who worked in the corporate setting, those who earned less, and those whose practice was more computerized expressed more concern about cost factors (dimension 1). These same characteristics, except for income, were also associated with a self-rated emphasis on monetary as opposed to nonmonetary cost items. However, none of the baseline characteristics was significantly related to whether respondents perceived physician-related vs. patient-related factors as important.

View this table:
Table 3

Between-cluster Characteristics for Barriers

1 (n1 = 269)2 (n2 = 236)3 (n3 = 114)4 (n4 = 69)Overall (n = 688*)
Mean values (SD)
Key dimensions
   1. Cost vs. noncost factors0.70 (0.19)0.06 (0.35)0.11 (0.24)0.23 (0.17)0.34 (0.39)
   2. Physician-related vs patient-related factors0.05 (0.32)0.18 (0.35)0.33 (0.27)–0.62 (0.11)0.07 (0.40)
   3. Monetary vs nonmonetary costs–0.07 (0.24)0.45 (0.21)–0.34 (0.22)0.09 (0.34)0.08 (0.38)
Explanatory variablesNo. (%)
   Male217 (81.0)189 (80.8)86 (76.1)58 (84.1)550 (80.4)
Years of work experience
   0–1079 (29.4)46 (19.6)36 (32.4)16 (23.2)177 (25.9)
   11–20115 (42.8)92 (39.2)29 (26.1)24 (34.8)260 (38.0)
   21–3056 (20.8)55 (23.4)24 (21.6)19 (27.5)154 (22.5)
   >3019 (7.1)42 (17.9)22 (19.8)10 (14.5)93 (13.6)
Work setting
   Corporate180 (71.4)116 (52.0)71 (68.3)42 (67.7)409 (63.8)
Monthly income (HK$)
   <100,00093 (35.5)76 (33.0)44 (39.3)15 (22.7)228 (34.0)
   100,000–149,99989 (34.0)71 (30.9)36 (32.1)27 (40.9)223 (33.3)
   150,000–199,99943 (16.4)28 (12.2)13 (11.6)9 (13.6)93 (13.9)
   >200,00037 (14.1)55 (23.9)19 (17.0)15 (22.7)126 (18.8)
Qualified specialist
   Yes157 (61.1)135 (62.8)63 (59.4)35 (56.5)390 (60.9)
Mean (SD)
No. of clinical functions computerized2.90 (2.39)2.32 (2.28)2.98 (2.32)3.25 (2.31)2.68 (2.37)
No. of administrative functions computerized2.87 (2.30)2.38 (2.21)2.87 (2.31)3.19 (2.26)2.68 (2.29)
P value
Explanatory variablesOverallDimension 1 (Clusters 1 and 2)Dimension 2 (Clusters 3 and 4)Dimension 3 (Clusters 2 and 3)
Years of work experience0.001<0.0010.300.03
Work setting<0.001<0.0010.940.006
Monthly income (HK$)
Qualified specialist0.820.710.710.56
No. of clinical functions computerized0.0040.0050.460.01
No. of administrative functions computerized0.020.020.360.06

SD = standard deviation.

  • * Does not equal the total sample size of 731 due to missing values.

  • Refers to the mean dimension scores on the biplots (Figures 1 and 2), in which a value of zero indicates no effect and progression in either the positive or negative directions indicates increasing effects.

  • Compares baseline explanatory variables across the two clusters that demonstrate the most extreme mean values for each key dimension, thereby characterizing the respondent characteristics most associated with each key dimension.

Similarly, three clusters resulted for the incentives/ catalysts variables (Table 4). Specialist qualifications were significantly linked with internal vs. external factors, and seniority and work setting were related to “push” vs. “pull” factors.

View this table:
Table 4

Between-cluster Characteristics for Incentives/Catalysts

1 (n1 = 392)2 (n2 = 153)3 (n3 = 139)Overall (n = 684*)
Mean values (SD)
Key dimensions
   1. Internal vs external factors0.68 (0.16)–0.12 (0.34)0.06 (0.27)0.38 (0.43)
   2. “Push” vs “pull” factors0.05 (0.35)0.43 (0.25)–0.45 (0.24)0.03 (0.42)
Explanatory variablesNo. (%)
   Male312 (80.2)118 (77.1)116 (84.1)546 (80.1)
Years of work experience
   0–10101 (25.9)46 (30.3)29 (20.9)176 (25.8)
   11–20139 (35.6)63 (41.5)57 (41.0)259 (38.0)
   21–3094 (24.1)22 (14.5)37 (26.6)153 (22.5)
   >3056 (14.4)21 (13.8)16 (11.5)93 (13.7)
Work setting
   Corporate251 (68.2)98 (69.0)59 (46.5)408 (64.1)
Monthly income (HK$)
   <100,000126 (32.9)55 (37.2)47 (35.1)228 (34.3)
   100,000–149,999126 (32.9)56 (37.8)38 (28.4)220 (33.1)
   150,000–199,99960 (15.7)15 (10.1)18 (13.4)93 (14.0)
   >200,00071 (18.5)22 (14.9)31 (23.1)124 (18.7)
Qualified specialist
   Yes237 (64.8)77 (54.2)74 (57.4)388 (60.9)
Mean (SD)
No. of clinical functions computerized2.86 (2.30)2.84 (2.45)2.29 (2.35)2.68 (2.37)
No. of administrative functions computerized2.91 (2.26)2.56 (2.28)2.38 (2.28)2.68 (2.29)
P value
Explanatory variablesOverallDimension 1 (Clusters 1 and 2)Dimension 2 (Clusters 2 and 3)
Years of work experience0.110.090.04
Work setting<0.0010.86<0.001
Monthly income (HK$)
Qualified specialist0.060.030.60
No. of clinical functions computerized0.040.930.06
No. of administrative functions computerized0.040.110.50

SD = standard deviation.

  • * Does not equal the total sample size of 731 due to missing values.

  • Refers to the mean dimension scores on the biplots (Figures 1 and 2), in which a value of zero indicates no effect and progression in either the positive or negative directions indicates increasing effects.

  • Compares baseline explanatory variables across the two clusters that demonstrate the most extreme mean values for each key dimension, thereby characterizing the respondent characteristics most associated with each key dimension.

Canonical Correlation

Table 5 shows correlation coefficients between barriers or incentives/catalysts with the two sets of derived canonical variables, B1/I1 and B2/I2. Panel A indicates that B1 correlates highly with capital costs, maintenance costs and lack of benefits, whereas B2 is linked to interference with communication. Correspondingly, in panel B it appears that financial incentives and government regulation are highly correlated with I1 and improved quality care with I2. Panel C reveals that respondents who were concerned with capital and maintenance costs and lack of benefits also perceived financial incentives and government regulation to be important incentives/ catalysts towards computerization. Similarly, those who found the potential interference with communication important also believed that the promise of improved quality care from computerization was a significant incentive/catalyst.

View this table:
Table 5

Correlation Coefficients of Barriers (Incentives/Catalysts) and Their Canonical Variables

ACanonical variates of barriersBCanonical variates of incentives/catalysts
1. Confidentiality breach0.18−0.291. Competitive pressures−0.090.18
2. Capital costs0.690.352. Financial incentives0.760.53
3. Interference with communication−0.24−0.533. Government regulation−0.480.36
4. Knowledge gap−0.100.224. Improved efficiency0.17−0.41
5. Lack of benefits−0.740.435. Improved quality care0.15−0.84
6. Lack of technical support−0.01−0.086. Patient demand−0.29−0.17
7. Maintenance costs0.440.017. Increased savings−0.320.28
8. Intellectual property regulations−0.090.36
9. Time costs−0.24−0.37
CCanonical variates of incentives/catalysts
1. Confidentiality breach0.05−0.05
2. Capital cost0.190.06
3. Interference with communication−0.06−0.10
4. Knowledge gap−0.030.04
5. Lack of benefits−0.200.08
6. Lack of technical support−0.004−0.01
7. Maintenance costs0.120.002
8. Intellectual property regulations−0.020.07
9. Time costs−0.07−0.07

Note: (B1, I1) and (B2, I2) are canonical pairs that were found to be statistically significant and demonstrated a satisfactory degree of correlation, where B1 = –0.76b1 – 0.52b2 – 1.02b3 – 0.92b4 – 1.44b5 – 0.89b6 – 0.57b7 – 0.64b8 – 1.12b9, I1 = 0.33i1 + 1.09i2 – 0.25i3 + 0.37i4 + 0.43i5 + 0.17i6 (sample canonical correlation = 0.27, p < 0.001); and B2 = –6.20b1 – 5.92b2 – 6.45b3 – 6.07b4 – 4.86b5 – 5.49b6 – 5.30b7 – 3.27b8 – 5.72b9, I2 = –0.20i1 + 0.01i2 – 0.15i3 – 0.50i4 – 0.96i5 – 0.45i6 (sample canonical correlation = 0.18, p = 0.04).


    This population-based physician survey systematically documented contributory barriers and potential incentives or catalysts that may influence clinical computerization in Hong Kong from the perspective of end-users. It provided empirical evidence regarding common barriers associated with information technology penetration. Moreover, the findings identified possible incentive strategies or catalyst schemes that may be employed to accelerate uptake of computer systems in the clinical setting, overcoming potential barriers and obstacles. The study's generalizability to the general physician population of Hong Kong lends further strength to the findings.

    Barriers to Computerization

    We found that barriers to computerization as perceived by physicians can generally be characterized by two types of factors—cost-related items and knowledge and attitudes of end-users regarding physician-related or patient-related items. The former can be further classified according to whether the costs are monetary, in terms of capital or maintenance charges, or more intangible items such as time and effort. The latter category refers to the mindset of end-users or decision-makers as they perceive the prerequisites to or consequences of a computerized practice. For instance, respondents in cluster 4 were most concerned with the knowledge and know-how required of themselves to implement and use computer systems (Table 3). There were also subgroups of physicians (clusters 2 and 3) who cared much about the effects of computerization on information privacy and interference with doctor-patient communication.

    Having identified these common barriers to computerization and recognizing the inherent heterogeneity of the general physician population, we looked for associated predictors of certain barriers for different subgroups of doctors. Baseline demographics and the practice environment were able to explain a large part of the cost-related items (dimensions 1 and 3), as would be expected.2,56 However, these same characteristics were not able to predict the knowledge/attitude axis (dimension 2) of the barrier variables. This was not surprising, given that we collected information on, and therefore were able to analyze, only sociodemographics and practice characteristics but not psychometric measures such as attitudinal indices, which would probably be more predictive and informative for this set of variables.

    Incentives for Computerization

    As for the range of potential incentives, we deduced two key dimensions explaining possible strategies to overcome barriers to computerization. First, the “pull” vs. “push” factors are the traditional “carrot-and-stick” approach to modifying behavior. Second, the within-practice vs. external or governmental factors represent an orthogonal but complementary approach to the previous axis of incentives. For example, the “pull” and within-practice factors can be translated into efficiency or financial gain incentives, for which the free market approach can be adopted to encourage clinical computerization through subsidies, tax credits, vouchers, or patient choice. When physician income has steadily declined by a considerable amount locally and it may be difficult to increase revenue or clinical throughput of patients, such efficiency gains in overhead expenses offered by clinical computerization should appeal to doctors who would like a resumption of their previous level of net remuneration. Alternatively or in combination, legislation and regulatory efforts form the other platform of catalyst schemes to increase the degree of computerization in clinical practice. This can be tied in with the enhanced clinical governance process demanded by medical boards and governments everywhere. For instance in the United Kingdom, the new Commission for Healthcare Audit and Inspection (CHAI) has recently been formed to oversee inspecting and licensing of medical facilities.13 There can be a defined set of minimum information technology requirements with which all medical practitioners must comply. Furthermore, in Ontario, Canada, where the provincial government pays all medical bills, physicians are required to file fee claims electronically.

    From the canonical correlation exercise, we can observe that tailored strategies can and should be devised for different subgroups of physicians in different communities depending on their perceived barriers and needs. For example, personal detailing sessions, successfully applied by pharmaceutical companies for years, that demonstrate the promise of improved quality of care from clinical computerization could be targeted at physicians who have expressed concerns about computers' potential interference with doctor-patient communication. On a community level, jurisdictions with a large proportion of doctors who are concerned about the cost implications of computerization (such as Hong Kong) could consider financial incentives or government regulation to help overcome these barriers to implementing computer systems in clinical practice.

    Potential Study Limitations

    Three important caveats of the study deserve mention. The cluster analytic technique, unlike traditional regressions, is not designed to control and adjust for multiple confounders. Therefore, the associations and predictors identified may not represent entirely independent relationships, although in our case this problem was not a major concern. Our primary purpose was to gain an understanding of the heterogeneity of responses and to link possible predictors to particular barriers or incentives respondents most valued. We did not set out to produce odds ratios or exact numerical estimates of the predictive effect size. In addition, we acknowledge that our findings may not be generalizable to and generic for all health systems. However, we suggest that the results are most relevant to jurisdictions with similar socioeconomic development (Singapore, Shanghai, Taipei), cultural identity (Beijing, Guangzhou, and other developed cities in China), and historical background (Singapore, Sri Lanka, India, and other former British colonies with similar health systems) to Hong Kong. On the other hand, we have little reason to believe the findings are applicable only in these locations. Further studies are needed to confirm the external validity of our results in different communities. Finally, future research, in which the questionnaire is applied across different populations and longitudinally (or repeatedly) in the same group of respondents, should aim to confirm the reliability and validity of the survey instrument.

    Conclusions and Recommendations

    Health systems of today place much emphasis on quality outcomes and cost reduction. Tools that allow greater accountability and evaluation represent a means for managers and policymakers to exercise more control over health care services to achieve these goals. Proper and swift implementation of computer systems in clinical practice can be seen as a prominent part of this overall philosophy. We have extended previous research findings by providing empirical support for the conceptualization of barriers toward computerization and linking these with corresponding incentive factors. The study also recognized and accounted for the heterogeneity of the general physician population by examining the results according to specific subgroups with different personal and practice profiles. Future research should focus on building implementation plans based on the present findings through overcoming the identified barriers and designing tailored incentive schemes for physicians with different characteristics and needs to encourage computerization, especially in the ambulatory solo and small group practice settings.

    Appendix: Questions Regarding Barriers and Incentives/Catalysts from the Survey Instrument

    Please rank the following

    Q3 The following factors have been identified as potential barriers to the computerisation of clinical practice.

    Please rank the top five factors (from the most to least important, no ties allowed) by completing the boxes below.


    1. Increased potential for breach of confidentiality of patient data

    2. Initial capital (monetary) cost of computerising a medical practice

    3. Interference with the nature and pattern of doctor-patient communication

    4. Lack of knowledge and perceived difficulty in learning new technology

    5. Lack of perceived benefits from computerisation of clinical practice

    6. Lack of technical support in case of system failure

    7. Maintenance costs (monetary) including computer, software upgrades and staff training

    8. Recent amendment of the Intellectual Property Ordinance to prevent copyright piracy

    9. Time costs associated with computerisation including planning, purchasing, training and maintenance

    Please rank the following

    Q4 The following factors have been identified as potential catalysts/incentives for the computerisation of clinical practice.

    Please rank the top five factors (from most to least important, no ties allowed) by completing the boxes below.


    1. Competitive peer pressure in terms of more practices becoming computerised

    2. Government financial incentives to support the computerisation of medical practices

    3. Government regulation requiring mandatory reporting of patient information

    4. Improved efficiency with real-time electronic access to laboratory results and hospital inpatient data

    5. Improved quality care through better access to medical evidence and information electronically

    6. Increased public or patient demand for a portable electronic medical record and more generally for a computerised practice

    7. Increased savings from efficiency gains in office operations (e.g. billing & inventory control)


    • The study received financial support from a “low budget-high impact” grant of the University of Hong Kong. We thank Marie Chi for expert secretarial assistance in the preparation of the manuscript.


    View Abstract