Abstract
Study objective
The impact of isolated gatekeeping on health care costs remains unclear. The aim of this study was to assess to what extent lower costs in a gatekeeping plan compared with a fee for service plan were attributable to more efficient resource management, or explained by risk selection.
Design
Year 2000 costs to the Swiss statutory sick funds and potentially relevant covariates were assessed retrospectively from beneficiaries participating in an observational study, their primary care physicians, and insurance companies. To adjust for case mix, two‐part regression models of health care costs were fitted, consisting of logistic models of any costs occurring, and of generalised linear models of the amount of costs in persons with non‐zero costs. Complementary data sources were used to identify selection effects.
Setting
A gatekeeping plan introduced in 1997 and a fee for service plan, in Aarau, Switzerland.
Participants
Of each plan, 905 randomly selected adult beneficiaries were invited. The overall participation rate was 39%, but was unevenly distributed between plans.
Main results
The characteristics of gatekeeping and fee for service beneficiaries were largely similar. Unadjusted total costs per person were Sw fr231 (8%) lower in the gatekeeping group. After multivariate adjustment, the estimated cost savings achieved by replacing fee for service based health insurance with gatekeeping in the source population amounted to Sw fr403–517 (15%–19%) per person. Some selection effects were detected but did not substantially influence this result. An impact of non‐detected selection effects cannot be ruled out.
Conclusions
This study hints at substantial cost savings through gatekeeping that are not attributable to mere risk selection.
Keywords: economics, health care costs, managed care programmes, gatekeeping
In the early 1990s Switzerland was among the first European countries to introduce managed care solutions.1 Health insurance is mandatory in Switzerland and these solutions were offered to the population as an alternative to traditional fee for service plans. In 2001, managed care organisations had a market share of 5%.2 Unlike in the USA, Swiss managed care lacks strong incentives to restrict the consumption of medical services.3,4
Early efforts to evaluate the medical and financial impact of managed care in Switzerland indicated reduced costs, but case mix adjustment was incomplete.5,6,7
Swiss gatekeeping plans report cost savings of 10%–25% compared with fee for service based health insurance.8 It remains unclear to what extent these savings are independent of risk selection mechanisms. Various studies have tried to answer similar questions for the USA and Europe, but findings were ambiguous.9,10,11,12,13,14,15
This study compares two local health plans, a gatekeeping and a fee for service plan, offered by the same group of health insurance companies in Aarau, Switzerland. These companies report costs to be about 10% lower in the gatekeeping plan, after adjusting for age and sex. We sought to assess to what extent this difference is attributable to more efficient resource management, or can be explained by risk selection.
Methods
Health plans
In the region of Aarau, a group of four companies provides health insurance to about 31 250 fee for service beneficiaries and 12 500 gatekeeping beneficiaries. The terms of fee for service insurance are uniformly defined by Swiss law. Free access to primary care physicians and medical specialists is guaranteed. The gatekeeping plan is managed by a single intermediary company. Its beneficiaries pay reduced insurance premiums. They are required to choose a primary care physician who will also act as a care coordinator and help avoiding unnecessary use of medical resources such as duplicate diagnostic tests. Specialist visits, except in emergencies, require referral by that coordinating physician. However, there is direct access to ophthalmological and gynaecological care. General coverage of medical services does not differ between plans. Gatekeeping physicians have no financial incentives to limit the use of medical services. They receive a minor administrative fee of Sw fr12 per inscribed patient per year. Any additional time spent on their coordinating function is reimbursed at normal rates—that is, it is reflected in the gatekeeping beneficiaries' cost to the Swiss statutory sick funds. Gatekeeping physicians have varying proportions of fee for service patients.
Study population
The population studied were 18 years or older in 1996 and either fee for service or gatekeeping beneficiaries throughout 2000, whether they consumed medical services or not. (According to the intermediary company managing the gatekeeping plan, there were hardly any beneficiaries who switched between health plans during the year, except for persons who moved into or out of the area.) Cohorts of 905 beneficiaries of each plan were randomly selected from the enrolment files. In early 2001, they were mailed an informed consent form including a self administered questionnaire. Only beneficiaries returning the questionnaire became known to the investigators. Further data were provided on consenting participants by their insurance companies and physicians. Three weeks after the first letter was sent, non‐responders were mailed a reminder. Data collection was completed in June 2001. Elaborate procedures were applied to ensure a maximum of data protection.
Study outcome and covariates
The primary outcome was gross cost to the Swiss statutory sick funds in 2000. Collection of covariates aimed to permit comprehensive case mix adjustment.16 It included cost and morbidity data between January and December 1996, the year before the gatekeeping plan was first offered to the population. Cost data and health insurance contract details were provided by the insurance companies, the latter reflecting the beneficiaries' decisions taken in 1999 and defining their status in 2000. The physicians provided morbidity scores for 1996 and 2000 (index of co‐existent diseases, ICED),17 physiological data, and their own practice characteristics. Data collection from the study participants comprised demographic and socioeconomic covariates; subjective health status (self administered SF‐36); health behaviour; inclination to overuse or underuse medical services; and medical resource use. As discussed later, some of these covariates were time dependent and reported the situation at the time of data collection—that is, in the first half of 2001. Analyses were performed including and excluding these covariates.
In addition to the main dataset, anonymous age and sex data for all randomly selected potential participants and year 2000 cost data for the total source population, aggregated by age, sex, and health plan, were available. These were used to identify selection effects.
All costs are in Swiss francs (Sw fr). On 31 December 2000, Sw fr1 equalled €0.66.
Statistical methods
Multiple logistic regression was used to model plan membership as a function of beneficiary characteristics identified in univariate analysis.
Expectedly, health care costs included a substantial proportion of zero values and were heavily left skewed and heteroskedastic (their variance increasing with increasing cost). Two‐part regression models of total and outpatient costs were fitted. In the first part we modelled whether any costs were accrued using logistic regression and in the second, generalised linear models (GLMs) were used to analyse the amount of costs in the persons with non‐zero costs.18,19 The GLMs used a logarithmic link function and assumed a γ distribution of errors.20,21 Potential covariates were assessed if an association with costs seemed plausible on logical or on statistical grounds (p⩽0.25 in univariate analysis). Firstly, all time dependent covariates primarily describing the situation in 2000 or 2001 were excluded (reduced models). In a second step (extended models), such covariates were permitted. Resource use variables were not used as covariates. As detailed in tables 4 and 5, some few observations with costs over Sw fr20 000 in 2000 were excluded from the main analysis, to reduce the impact of chance effects in this small sample. Complementary analyses included all available observations.
Table 4 Generalised linear model of total health care costs in 2000, per person with non‐zero costs* (part 2 of two part model).
| Number = 347 | Log‐likelihood −2988.97 | Deviance 249.15 |
|---|---|---|
| Independent variable | Coeff (95% CI) | p Value |
| Fee for service plan membership | 0.24 (0.04 to 0.44) | 0.021 |
| Age (women) | −0.03 (−0.09 to 0.02) | 0.246 |
| Age2 (women) | 0.001 (0.000 to 0.001) | 0.007 |
| Age (men) | −0.13 (−0.21 to −0.04) | 0.003 |
| Age2 (men) | 0.003 (0.001 to 0.005) | 0.002 |
| Age3 (women) | −0.000 (−0.000 to −0.000) | 0.007 |
| 1996 outpatient costs (log scale) | 0.06 (0.01 to 0.10) | 0.010 |
| ICED score in 1996 | 0.61 (0.28 to 0.94) | <0.001 |
| ICED score in 1996 * age (female)† | −0.01 (−0.01 to −0.00) | <0.001 |
| ICED score in 1996 * age (male)† | −0.01 (−0.01 to −0.00) | <0.001 |
| ICED score in 1996 * SF‐36 Item 2† | −0.07 (−0.12 to −0.01) | 0.025 |
| ICED score in 1996 * 1996 outpatient costs (log scale)† | 0.02 (0.00 to 0.05) | 0.040 |
| ICED score increase between 1996 and 2000 | 0.13 (0.06 to 0.20) | 0.001 |
| SF‐36 Item 2 | −0.13 (−0.29 to 0.03) | 0.121 |
| SF‐36 General Health Scale score | −0.01 (−0.02 to −0.01) | <0.001 |
| Complementary semi‐private insurance | −0.29 (−0.50 to −0.09) | 0.005 |
| Importance assigned to low insurance premiums‡ | 0.10 (0.02 to 0.19) | 0.017 |
| Living in a partnership | 0.60 (0.19 to 1.01) | 0.004 |
| Marital status | ||
| married§ | 0.27 (−0.15 to 0.68) | 0.208 |
| widowed§ | 0.00 (−0.48 to 0.49) | 0.989 |
| divorced§ | −0.63 (−1.16 to −0.10) | 0.19 |
| Household size | ||
| 2 adults¶ | −1.25 (−1.67 to −0.82) | <0.001 |
| ⩾2 adults¶ | −1.39 (−1.88 to −0.90) | <0.001 |
| Integration | ||
| Swiss born or Swiss citizen** | −0.39 (−0.88 to 0.11) | 0.127 |
| Swiss born and Swiss citizen** | 0.11 (−0.33 to 0.54) | 0.624 |
| Aarau area residency in 1996 | 2.04 (1.14 to 2.95) | <0.001 |
| Aarau area residency in 1996 * age (female) † | −0.02 (−0.05 to 0.00) | 0.069 |
| Constant | 7.36 (5.69 to 9.03) | <0.001 |
*Three gatekeeping observations and four fee for service observations with health care costs over Sw fr20000 in 2000 not used. †Term representing effect modification. ‡Per increase by 1 on a 5 point Likert scale. §Compared with unmarried. Combined likelihood ratio test, p = 0.054. ¶Compared with one adult. **Compared with neither Swiss born nor Swiss citizen. Combined likelihood ratio test, p = 0.034.
Table 5 Estimated cost impact of gatekeeping plan membership compared with fee for service plan membership (based on two part models).
| Number* | Cost difference (fee for service minus gatekeeping) in persons with non‐zero costs‡ | Number† | Cost difference (fee for service minus gatekeeping) in all persons§ | Cost difference (fee for service minus gatekeeping) in all persons (%)§¶ | |
|---|---|---|---|---|---|
| Total costs | |||||
| Reduced model | 372 | 498 (−77 to 1072) | 439 | 403 (−120 to 1027) | 14.5 |
| Extended model | 347 | 513 (53 to 973) | 395 | 517 (−11 to 1254) | 18.9 |
| Outpatient costs | |||||
| Reduced model | 377 | 544 (76 to 1014) | 444 | 453 (28 to 973) | 24.6 |
| Extended model | 354 | 394 (23 to 765) | 402 | 372 (−4 to 813) | 17.9 |
*Number available for GLM fitting. Reduced and extended total cost models, three gatekeeping observations and four fee for service observations with health care costs over Sw fr20000 in 2000 not used. Reduced outpatient cost model, one fee for service observation with outpatient costs over Sw fr20000 in 2000 not used. In the extended outpatient cost model, this observation was not contained because of a missing value in one of the additional predictor variables. †Number available for estimation of marginal (population) effects. ‡Conditional effect in persons with non‐zero costs as derived from GLM coefficients. Mean Sw fr per person (CI). §Marginal (population) effect (combined effect estimate of two part regression, comparing the assumptions of exclusive gatekeeping plan membership and with exclusive fee for service plan membership). Mean Sw fr per person (bootstrapped CI). ¶Expressed as a percentage of the costs incurred by the fee for service source population in the year of reference.
Total predicted values were calculated by multiplying the predicted values of both sub‐models.21 To estimate the marginal (population level) cost impact of gatekeeping, all participants were assumed to be gatekeeping beneficiaries, or fee for service beneficiaries. Both sets of predicted values were calculated and their difference was taken. The result estimates the cost impact of replacing fee for service based health insurance with gatekeeping in the source population.
Two sided p values of 0.05 were used to determine significance. Confidence intervals (CIs) shown are at the 95% level. CIs for the marginal effects were calculated by bias corrected bootstrapping using 1000 repetitions.
Results
Participation and data availability
In total, 700 (39%) of the randomly selected persons returned the mailed questionnaire, 433 (48%) of the gatekeeping beneficiaries and 267 (30%) of the fee for service beneficiaries. In both groups, 86% of these consented to have additional data collected from their insurance companies and physicians. Full data inclusive of year 2000 cost data and year 1996 cost and morbidity data were finally available from 466 (26%) of the randomly selected persons, 317 (35%) of the gatekeeping beneficiaries and 149 (16%) of the fee for service beneficiaries.
Data completeness among respondents was at least 90%. Data provided by a total of 82 participating physicians were near complete, but some physicians who only treated fee for service beneficiaries refused to participate, which reduced the number of fee for service beneficiaries with full data available. Data provided by the insurance companies were complete.
Beneficiary characteristics and health status
Demographic characteristics, health insurance contract details, and indicators of socioeconomic status were similar between plans (table 1). The age range was 23–92 years in the gatekeeping group and 23–96 years in the fee for service group. However, the gatekeeping beneficiaries were on average 3.2 years older than the fee for service beneficiaries and the proportion of women was lower by 7%. The gatekeeping beneficiaries seemed to be slightly less mobile, less professionally active, and had a lower household income.
Table 1 Selected beneficiary characteristics by plan.
| Characteristic | Gatekeeping (n = 433)* | Fee for service (n = 267)* | p Value |
|---|---|---|---|
| Age (mean (SD) years) | 56.8 (17.1) | 53.6 (16.3) | 0.014¶ |
| Female (%) | 53.2 | 60.3 | 0.068** |
| Duration of health insurance with the same company (mean (SD) years) | 29.7 (18.2) | 28.8 (18.0) | 0.555¶ |
| Complementary insurance contracts (mean (SD) number) | 0.94 (0.75) | 1.0 (0.70) | 0.088 |
| Importance assigned to low insurance premiums (mean (SD) score)† | 2.8 (1.1) | 2.5 (1.2) | 0.003†† |
| Professional status (%)‡ | |||
| professionally active | 48.7 | 60.1 | 0.004** |
| housework and childcare | 47.1 | 52.4 | 0.171** |
| unemployed | 1.9 | 2.6 | 0.492** |
| retired | 41.6 | 32.2 | 0.013** |
| Marital status (%) | |||
| unmarried | 10.0 | 16.7 | 0.055** |
| married | 71.6 | 64.0 | |
| widowed | 11.2 | 10.6 | |
| divorced | 7.2 | 8.7 | |
| Household size (mean (SD) number) | |||
| adults | 2.0 (0.7) | 1.9 (0.8) | 0.208†† |
| children ⩽ 18 years | 0.40 (0.82) | 0.35 (0.73) | 0.631†† |
| Nursing home residency (%) | 1.9 | 1.9 | 0.988** |
| Residency in the Aarau area in 1996 (%) | 97.2 | 90.6 | <0.001** |
| BMI (mean (SD) kg/m2) | 25.2 (4.4) | 24.9 (4.4) | 0.498¶ |
| Physically active or doing sports (%) | 38.2 | 44.3 | 0.113** |
| Current smoking (%) | 22.2 | 30.0 | 0.022** |
| Current alcohol consumption | 87.0 | 87.2 | 0.801** |
| Importance assigned to healthy nutrition (mean (SD) score)§ | 3.09 (0.67) | 3.04 (0.62) | 0.267†† |
| Subjective health status (mean (SD) score) | |||
| SF‐36 general health scale | 70.6 (18.5) | 71.3 (19.9) | 0.648¶ |
| SF‐36 physical health summary scale | 49.6 (9.9) | 50.0 (10.2) | 0.648¶ |
| SF‐36 mental health summary scale | 52.1 (8.9) | 51.2 (9.4) | 0.247¶ |
| ICED (mean (SD) score) | |||
| in 2000 | 2.2 (3.1) | 2.0 (2.9) | 0.470†† |
| in 1996 | 1.8 (2.7) | 1.5 (2.4) | 0.378†† |
| History of mental illness (%) | 17.1 | 18.8 | 0.635** |
*Number is slightly smaller at the individual variable level because of missing values. †Score on a 5 point Likert scale. ‡Several answers could be ticked. §Score on a four point Likert scale. ¶Unpaired t test. **χ2 test. ††Mann‐Whitney U test.
Both groups were similar with respect to health behaviour and health status (table 1). However, the gatekeeping group had a lower proportion of current smokers, especially in the younger age groups. The proportion of physically active persons was higher in the fee for service group (non‐significant).
Medical resource use and cost
Table 2 details medical resource use and cost to the Swiss statutory sick funds. Fewer consultations with medical specialists and fewer hospitalisations were reported in the gatekeeping group and their year 2000 total costs per person were Sw fr231 lower. Outpatient costs were Sw fr377 lower (consultation costs, Sw fr7 lower; medication costs, Sw fr130 lower; other outpatient costs Sw fr239 lower). Inpatient costs were Sw fr145 higher in the gatekeeping group, but this difference was annulled when nine observations with costs over Sw fr20 000 were excluded. Year 1996 costs were Sw fr762 lower in the gatekeeping group, but this difference was reduced to Sw fr199 when six observations with costs over Sw fr20 000 were excluded. None of the differences observed were statistically significant.
Table 2 Resource use and cost to the Swiss statutory sick funds by plan.
| Characteristic | Gatekeeping (n = 433)* | Fee for service (n = 267)* | p Value§ |
|---|---|---|---|
| Primary care physician consultations in 2000† | 3.2 (4.4); 2 | 3.2 (4.2); 2 | 0.440 |
| Medical specialist consultations in 2000† | 1.0 (2.2); 0 | 1.6 (4.0); 0 | 0.083 |
| Hospitalisations in 2000† | 0.16 (0.44); 0 | 0.21 (0.58); 0 | 0.664 |
| Total costs > Sw fr0 in 2000 (%) | 83.6 | 83.6 | 0.984 |
| Total costs in 2000‡ | 2496 (4870); 1120 | 2727 (4311); 1344 | 0.407 |
| Outpatient costs in 2000‡ | 1815 (2287); 1102 | 2192 (3113); 1261 | 0.382 |
| Inpatient costs in 2000‡ | 680 (3821); 0 | 535 (1948); 0 | 0.994 |
| Total costs > Sw fr0 in 1996 (%) | 80.7 | 79.1 | 0.634 |
| Total costs in 1996‡ | 1674 (2991); 731 | 2436 (5466); 824 | 0.646 |
| Outpatient costs in 1996‡ | 1284 (1702); 670 | 1648 (2726); 811 | 0.469 |
| Inpatient costs in 1996‡ | 390 (1934); 0 | 789 (3849); 0 | 0.375 |
*Number is smaller at the individual variable level because of missing values. †Mean number (SD); median. Self reported values, in good accordance with physician reported values. ‡Mean Sw fr (SD); median. Observations with zero values included. §Mann‐Whitney U test.
Complementary data sources
Comparison with complementary data sources was undertaken to identify selection effects. Among all randomly selected persons, the observation of a higher mean age and a lower proportion of women on the gatekeeping side was confirmed, but less distinct. Participation rates by age and sex group showed moderate deviations (1%–17%) from mean plan specific participation rates.
After adjusting for resulting differences in the age and gender distribution and after excluding all cases with costs over Sw fr20 000, year 2000 study level costs and the corresponding aggregated costs for the source population were similar. In the gatekeeping plan, study level total costs per person were Sw fr86 lower than population level costs, and in the fee for service plan, they were Sw fr69 lower. Within the strata defined by an age cut off of 65 years and sex, some of the differences seen were more distinct, but still moderate. The female fee for service beneficiaries above 65 years of age were the only exception. Their study level total costs were Sw fr740 lower than in the source population, compared with Sw fr8 lower in the corresponding gatekeeping beneficiaries.
Predictors of plan membership
Logistic regression indicated that gatekeeping plan membership in 2000 was positively associated with lower 1996 total health care costs; higher 1996 ICED score; having complementary dental insurance; having a higher importance assigned to healthy nutrition; having a lower household income; having more children in the household; living in the Aarau area in 1996; and having a primary care physician with a higher number of consultations per year. The explanatory power of the model remained low (pseudo r2 0.10, 71% correct predictions).
Predictors of cost
In the reduced model, non‐zero total costs in 2000 were associated with higher 1996 outpatient costs; higher 1996 ICED score; lower age (note: after correction for morbidity). The effect of plan membership was modified by age, hinting at a reduced probability of non‐zero costs in younger gatekeeping beneficiaries and vice versa. Table 3 shows the extended model.
Table 3 Logistic regression model of non‐zero total health care costs in 2000 (part 1 of two part model).
| n = 418* | Log‐likelihood −99.05† | Pseudo r2 of the model = 0.43‡ |
|---|---|---|
| Independent variable | Coeff (95% CI) | p Value |
| Fee for service plan membership | −3.43 (−5.76 to −1.12) | 0.004 |
| Fee for service plan membership divided by age (women)§ | 138.51 (35.37 to 241.65) | 0.008 |
| Fee for service plan membership divided by age (men)§ | 143.83 (41.24 to 246.42) | 0.006 |
| Age (women) | 0.01 (−0.08 to 0.11) | 0.762 |
| Age2 (women) | −0.001 (−0.002 to −0.000) | 0.020 |
| Age (men) | −0.06 (−0.12 to −0.01) | 0.021 |
| 1996 outpatient costs (log scale) | 0.31 (0.18 to 0.43) | <0.001 |
| ICED score in 1996 | 0.67 (0.12 to 1.22) | 0.017 |
| ICED score increase between 1996 and 2000 | 0.73 (0.14 to 1.31) | 0.016 |
| SF‐36 General Health Scale score | −0.03 (−0.06 to −0.01) | 0.018 |
| Fixed beneficiary co‐payment¶ | ||
| Sw fr400 | −0.85 (−1.76 to 0.06 | 0.067 |
| Sw fr600 | −0.92 (−2.23 to 0.39) | 0.171 |
| ⩾Sw fr1200 | −1.92 (−3.04 to −0.79) | 0.001 |
| Importance assigned to low insurance premiums** | 0.41 (0.07 to 0.75) | 0.019 |
| Self reported low aversion of consulting a doctor** | −0.43 (−0.81 to −0.06) | 0.024 |
| Being retired | 1.75 (0.35 to 3.14) | 0.014 |
| Constant | 5.42 (1.96 to 8.89) | 0.002 |
*Number<466 because of missing values. †Uncritical Hosmer‐Lemeshow goodness of fit test (p = 0.54). ‡Predictions correct in 89%. §Term representing effect modification. ¶Compared with lawful minimum of CHF 230. **Per increase by 1 on a 5 point Likert scale.
In the study of participants with non‐zero costs, the reduced GLM showed higher total costs to be significantly associated with fee for service plan membership (likelihood ratio test, borderline p = 0.066); higher 1996 outpatient costs; higher 1996 ICED score; higher age; choice of lower self‐payments but higher insurance premiums; having complementary semi‐private insurance; living in the Aarau area in 1996 (modified by age in women). The effect of 1996 ICED score was modified by age and 1996 outpatient costs. Table 4 shows the extended model.
Comparison of predicted and observed costs per person showed an overestimation in the fee for service group (difference Sw fr283 when regarding the extended total cost model), but not in the gatekeeping group (difference Sw fr−5). When the female fee for service beneficiaries above 65 years of age (17 observations) were excluded, the difference seen in the fee for service group was reduced to Sw fr−86.
Association of gatekeeping and cost
Direct parameter estimates derived from the conditional cost models as well as the estimated marginal (population) effects, comparing exclusive gatekeeping plan membership with exclusive fee for service plan membership, showed costs savings through gatekeeping at the total and outpatient levels (table 5). Estimated savings per person were in the range of Sw fr403–517 (15%–25% of the costs incurred by the fee for service source population). Some of the bootstrap based confidence intervals for the marginal effects overlapped the null, but there was a strong and uniform tendency towards savings by gatekeeping. The reduced and extended models yielded consistent results at the total costs level, but the effect estimate derived from the reduced outpatient cost model seemed high.
What this paper adds
The aim of gatekeeping is to reduce the cost of health care without affecting its quality, primarily by avoiding duplicate diagnostic tests and unnecessary consultations with specialists.
Studies of the impact of gatekeeping in mixed settings, where other techniques of utilisation management were also in place, have been inconclusive. According to this study, isolated gatekeeping may be an efficient technique of utilisation management.
Re‐fitting the models and recalculating the marginal effects after inclusion of up to seven observations with costs over Sw fr20 000 led to higher effect estimates (for example, Sw fr773 instead of Sw fr517 when using the extended total cost model). In contrast, decreasing the cut off point further to Sw fr15 000 changed the effect estimates only marginally (Sw fr486 instead of Sw fr517). Exclusion of the female fee for service beneficiaries above 65 years of age yielded higher effect estimates (Sw fr645 instead of Sw fr517). Exclusion of the persons who joined the gatekeeping plan later than in 1997 yielded results in the range of the main results (Sw fr481 instead of Sw fr517).
Discussion
This study of a gatekeeping and a fee for service plan in Aarau, Switzerland, hints at relevant cost savings through gatekeeping that are not attributable to mere risk selection. Adjustment for case mix was achieved by performing two part multivariate analyses of year 2000 costs to the Swiss statutory sick funds, taking into account a wide variety of beneficiary and physician characteristics. The characteristics of gatekeeping beneficiaries and fee for service beneficiaries were largely similar. A considerable difference in the proportion of current smokers was concentrated on the younger study participants where a substantial impact on health care costs would not yet be expected. Whether physicians treated fee for service beneficiaries only, or beneficiaries from both plans, was not a significant predictor of cost on the fee for service side.
The result of case mix adjusted gatekeeping associated savings of around 20% confirms earlier Swiss reports and earlier, mostly trial based findings from the USA that gatekeeping may be an efficient technique of utilisation management.9,10,11,22,23,24 However, non‐randomised US studies found no or only marginal costs savings associated with gatekeeping.13,14,25,26,27 A European study using country level aggregate data found no gatekeeping effect on total costs, but significant savings in the outpatient setting.15
In this study, constraints on planned sample size in conjunction with a low response rate and incomplete information from some participants led to a small number of usable observations. Thus, the power to detect differential plan member characteristics may have been limited. Furthermore, response rates differed considerably between plans, hinting at the possibility of selection bias.
Accounting for a wide range of potential confounders reduced the probability of strong selection bias. Moreover, external data allowed us to assess in part to what extent selection effects were present in the study dataset. Essentially, comparison with aggregated cost data for the source population showed that this study found low costs in the female fee for service beneficiaries from age 65 onwards.
Comparison of predicted and observed costs at the GLM level showed an isolated over‐estimation of costs in the fee for service group, implying a possible exaggeration of the gatekeeping effect. Exclusion of the observations representing female fee for service beneficiaries from age 65 onwards diluted this over‐estimation. Re‐estimation of the two part cost models after exclusion of this same group of observations did not reduce the combined estimates of the gatekeeping effect. This latter finding may suggest that the identified deviation of observed costs from population level costs at the subgroup level induced no strong distortion of the main study results. However, additional influences of non‐detected selection effects alongside unmeasured covariates cannot be ruled out.
Policy implications
Policy decisions that have an impact on national reimbursement systems, or on the type of products offered by health insurers, should consider gatekeeping as an option that avoids strong incentives to restrict the use of medical services and that may, therefore, be largely uncontroversial for a public concerned with quality.
Cost and morbidity data for 1996, the year before the gatekeeping plan was first offered to the population, defined a baseline that could not be influenced by plan specific mechanisms.
The finding of cost savings through gatekeeping is not invalidated by the observation that the unadjusted increase in total health care costs between 1996 and 2000 was more pronounced on the gatekeeping side. As confirmed by logistic regression modelling of plan membership, persons with higher health care costs in 1996 but not persons with a higher 1996 ICED score were reluctant to join the gatekeeping plan. This must have led to a regression to the mean effect as it has been described before for similar settings.7 A higher mean age in the gatekeeping group and the fact that more persons in this group reached the age threshold of 65 years between 1996 and 2000 may also have contributed to the more pronounced increase seen here.
Data collection from the study participants occurred in the first half of 2001 and some time dependent covariates (describing subjective health status, health behaviour, and resource use preferences) reported the situation at this point in time, but were nevertheless used in the regression models on year 2000 costs. The intention was not to assess cause‐effect relations between these covariates and the target variable, but merely to reduce the amount of unexplained variance and thus to achieve more precise estimates of the gatekeeping effect. Moreover, most of the covariates in question tend to change slowly over time. The differences between the health plans under study are fairly limited and unlikely to cause differential changes of attitudes, behaviours or even health status in the mid‐term. Thus, the 2001 values of these covariates can be assumed to represent the situation directly before and in 2000 fairly adequately, except for random changes of health status.
All regression analyses on cost were performed excluding as well as including these covariates and the resulting estimates of the gatekeeping effect were similar. However, the reduced model on outpatient costs yielded effect estimates that were higher than those seen in the total costs. This was because of insufficient adjustment and it turned out that the difference in explanatory power between the reduced and extended models occurred mainly because no subjective health status variable was available for the former. The level of self payments chosen (an insurance contract detail) tended to be higher in healthier persons and could be used as an, albeit unsatisfactory, proxy. Variables indicating a high importance assigned to low insurance premiums, or a low household income, behaved near identical in all cost models, but the former was less affected by missing values and therefore preferred.
The gatekeeping plan under study does not incorporate additional utilisation management practices such as prospective utilisation review.28 It focuses on the avoidance of duplicate diagnostic tests and unnecessary specialist consultations. Some case management occurs informally. The exact mechanisms behind the cost savings seen could not be identified, as cost and resource use data were not detailed enough for a refined analysis of the medical services provided. Therefore, we cannot contribute to ongoing discussions whether the gatekeeping approach could be optimised, for example, by permitting direct specialist access for particular subgroups of persons or under special circumstances.29,30 In our case, most savings were realised in the outpatient setting. The number of consultations was less important for the overall result than the amount of services performed per consultation and the amount of medications prescribed.
This study supports that utilisation management through gatekeeping may be associated with relevant savings in health care costs.
Acknowledgements
Beat Gasser, Atupri Krankenkasse, Berne; Asy Durrer, CONCORDIA Kranken‐ und Unfallversicherer, Lucerne; Dr René Kühne, Helsana Versicherungen AG, Zürich; Denise Fitzi, Winterthur Versicherungen, Winterthur; and Dr Arthur Vogt, Steinen, all Switzerland, contributed during preparation and conduction of the study. Dr Susan Ettner, UCLA Schools of Medicine and Public Health, Los Angeles, USA, and Professor Christoph Minder, ISPM, Berne, Switzerland, advised on statistical issues. Dr Adam Lowy, SIAK, Berne, Switzerland, and Professor Sebastian Schneeweiss, Boston, USA, made valuable comments on the manuscript.
Footnotes
Funding: this work was partially funded by unrestricted, educational grants from CONCORDIA Schweizerische Kranken‐ und Unfallversicherung, Lucerne; Helsana Versicherungen AG, Zürich; Krankenkasse SBB, Zürich; Wincare Versicherungen, Winterthur; Novartis Pharma Schweiz AG, Berne, all Switzerland.
Conflicts of interest: Georges Preiswerk is an employee of the company managing the gatekeeping plan under study. Roman Lehner and Fritz Weber are office based physicians treating beneficiaries of the health plans under study. Matthias Schwenkglenks and Thomas D Szucs: no potential conflicts of interest.
Ethical approval: On 27 February 2001, the data protection registrar of University Hospital, Zürich, Switzerland, the corresponding author's affiliation when this study was planned and data were collected, decided that an ethical approval was not required for this study as the design did not entail an intervention.
References
- 1.Bohlert I, Adam I, Robra B P. [The Swiss gatekeeper system ‐ a model for improving capacity development and economic effectiveness]. Gesundheitswesen 199759488–494. [PubMed] [Google Scholar]
- 2.Anonymous Managed‐Care‐Modelle in der Schweiz. Managed Care 2001537–39. [Google Scholar]
- 3.Kuttner R. Must good HMOs go bad? First of two parts: the commercialization of prepaid group health care, N Engl J Med 19983381558–1563. [DOI] [PubMed] [Google Scholar]
- 4.Baumberger J.So funktioniert Managed Care. Anspruch und Wirklichkeit der integrierten Gesundheitsversorgung in Europa. Stuttgart: Thieme, 2001126–132.
- 5.Bundesamt für Sozialversicherung Neue Formen der Krankenversicherung: Alters‐ und Kostenverteilungen im Vergleich zu der traditionellen Versicherung. Ergebnisse der Administrativdatenuntersuchung, 2. Teil. Bern: Bundesamt für Sozialversicherung, 199811136–141. [Google Scholar]
- 6.PROGNOS AG Evaluation neuer Formen der Krankenversicherung. Synthesebericht. Bern: Bundesamt für Sozialversicherung, 199839–58.
- 7.Etter J F, Perneger T V. Health care expenditures after introduction of a gatekeeper and a global budget in a Swiss health insurance plan. J Epidemiol Community Health 199852370–376. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Baumberger J.So funktioniert Managed Care. Anspruch und Wirklichkeit der integrierten Gesundheitsversorgung in Europa. Stuttgart: Thieme, 2001136–137.
- 9.Schillinger D, Bibbins‐Domingo K, Vranizan K.et al Effects of primary care coordination on public hospital patients. J Gen Intern Med 200015329–336. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Martin D P, Diehr P, Price K F.et al Effect of a gatekeeper plan on health services use and charges: a randomized trial. Am J Public Health 1989791628–1632. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Galt K A, Rich E C, Kralewski J E.et al Group practice strategies to manage pharmaceutical cost in an HMO network. Am J Manag Care 200171081–1090. [PubMed] [Google Scholar]
- 12.Hurley R E, Freund D A, Gage B J. Gatekeeper effects on patterns of physician use. J Fam Pract 199132167–174. [PubMed] [Google Scholar]
- 13.Bonham G S, Barber G M. Use of health care before and during Citicare. Med Care 198725111–119. [DOI] [PubMed] [Google Scholar]
- 14.Vertrees J C, Manton K G, Mitchell K C. Case‐mix adjusted analyses of service utilization for a Medicaid health insuring organization in Philadelphia. Med Care 198927397–411. [DOI] [PubMed] [Google Scholar]
- 15.Delnoij D, Van Merode G, Paulus A.et al Does general practitioner gatekeeping curb health care expenditure? J Health Serv Res Policy 2000522–26. [DOI] [PubMed] [Google Scholar]
- 16.Schneeweiss S, Sangha O. [Provider profiling: needs, methodologic requirements and means to increase acceptance]. Dtsch Med Wochenschr 2001126918–924. [DOI] [PubMed] [Google Scholar]
- 17.Greenfield S, Apolone G, McNeil B J.et al The importance of co‐existent disease in the occurrence of postoperative complications and one‐year recovery in patients undergoing total hip replacement. Comorbidity and outcomes after hip replacement. Med Care 199331141–154. [DOI] [PubMed] [Google Scholar]
- 18.Duan N, Manning W G J, Morris C N.et al A comparison of alternative models for the deman for medical care. Journal of Business and Economic Statistics 19831115–126. [Google Scholar]
- 19.Mullahy J. Much ado about two: reconsidering retransformation and the two‐part model in health econometrics. J Health Econ 199817247–281. [DOI] [PubMed] [Google Scholar]
- 20.Diehr P, Yanez D, Ash A.et al Methods for analyzing health care utilization and costs. Annu Rev Public Health 199920125–144. [DOI] [PubMed] [Google Scholar]
- 21.Blough D K, Madden C W, Hornbrook M C. Modeling risk using generalized linear models. J Health Econ 199918153–171. [DOI] [PubMed] [Google Scholar]
- 22.Huber‐Stemich F, Hees K, Baumann P.et al Sechs Jahre HMO Zürich‐Wiedikon. Ein Erfahrungsbericht. Ars Medici 1996181079–1082. [Google Scholar]
- 23.Frey W. HMO‐ und Hausarztmodelle in der Schweiz. KSK aktuell. Konkordat der Schweizerischen Krankenversicherer 1996454–55. [Google Scholar]
- 24.Manning W G, Leibowitz A, Goldberg G A.et al A controlled trial of the effect of a prepaid group practice on use of services. N Engl J Med 19843101505–1510. [DOI] [PubMed] [Google Scholar]
- 25.Hurley R E, Paul J E, Freund D A. Going into gatekeeping: an empirical assessment. QRB Qual Rev Bull 198915306–314. [DOI] [PubMed] [Google Scholar]
- 26.Long S H, Settle R F. An evaluation of Utah's primary care case management program for Medicaid recipients. Med Care 1988261021–1032. [DOI] [PubMed] [Google Scholar]
- 27.Grembowski D E, Martin D, Diehr P.et al Managed care, access to specialists, and outcomes among primary care patients with pain. Health Serv Res 2003381–19. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.Wickizer T M, Lessler D. Utilization management: issues, effects, and future prospects. Annu Rev Public Health 200223233–254. [DOI] [PubMed] [Google Scholar]
- 29.Ferris T G, Chang Y, Blumenthal D.et al Leaving gatekeeping behind—effects of opening access to specialists for adults in a health maintenance organization. N Engl J Med 20013451312–1317. [DOI] [PubMed] [Google Scholar]
- 30.Escarce J J, Kapur K, Joyce G F.et al Medical care expenditures under gatekeeper and point‐of‐service arrangements. Health Serv Res 2001361037–1057. [PMC free article] [PubMed] [Google Scholar]