Annexures to the report on the analysis of REF shadow returns

Transcription

1 Annexures to the report on the analysis of REF shadow returns Contents Annexure A: Annexure B: Changes introduced in Version 2. of the Entry & Verification Criteria during...9 Cluster analysis underlying the scheme-specific expected rates for CDLs...2 Annexure C: Category definitions...32 Annexure D: DIN Score methodology...33 Annexure E: Annexure F: REF submissions for, the categorisation thereof, and the potential financial impact of the REF...38 Details on REF risk factors where schemes have reported significant deviations from the expected...58 Annexure G: REF price by age curves and community rate analysis for administrator groups Private Bag X34, HATFIELD, 28 Hadefields Block E, 267 Pretorius Street, HATFIELD REF shadow returns 8

2 Annexure A: Changes introduced in Version 2. of the Entry & Verification Criteria during 2 April Due to the findings of the REF pricing study that has recently been completed and comments received from the industry, a number of changes to Version 2 of these guidelines has become necessary. Certain technical omissions have also been corrected. Changes are made to the following areas 4 : The month in which a beneficiary is counted is now based on service date and not on payment date It is now specified that authorisation is the only source for ICD diagnosis codes CDLs occurring in beneficiaries under one year of age can no longer be counted. These cases must be reflected in the NON-column The admission date is to be used to determine when a maternity event is recorded The ATC code, B2BD6 (Von Willebrand factor and coagulation factor VIII in combination) has been added as proof of treatment for haemophilia To effect the above, changes were made to the following sections: a. Clarification of ambiguous wording: Sections 3.9, 3.9., 3.5, b. Definition of service date for maternity Section 3.: c. The following sections have been altered to deal with the inclusion of all beneficiaries in the under one age-band in the NON column: Sections 3.3 (A), 3.6, 3.2, 3.3 d. The use of diagnoses obtained through authorisation is specified: Sections 5., 5.3(A), 5.3.3, 5.4.3, 5.9 e. Clarification that service date must be used to define a beneficiary s month for eligibility Sections 5.4.4, 5.6, all the Boolean tables in Section 6, f. Technical oversights H2AB (Glucocorticoids) have been removed from the Boolean tables for Asthma, COPD, and Multiple sclerosis Addition of the ATC code, B2BD6 (Von Willebrand factor and coagulation factor VIII in combination) for Haemophilia (Table 5 and ATC code descriptions in Section 7) The cost hierarchy of the respective CDL s are now presented in section Note that the hierarchy for respiratory conditions has changed (Section 3.9..) 4 This page is an excerpt from Guidelines for the Identification of Beneficiaries with REF Risk Factors in Accordance with the REF Entry and Verification Criteria, Version 2., 2 April. Available at: _REF_Risk_Factors_Version_2_.pdf REF shadow returns 9

3 Annexure B: Cluster analysis underlying the scheme-specific expected rates for CDLs Contents BACKGROUND INTRODUCTION TO CLUSTER ANALYSIS Clustering methods Hierarchical Cluster Analysis K-Means Cluster Analysis Similarity of - and distance between clusters Average Centroid Ward Evaluation criteria applied in evaluating clustering models Pseudo-F statistic Cubic Clustering criterion Approximate Overall R APPLICATION OF CLUSTERING TECHNIQUES TO BENEFIT OPTIONS The data Clustering variables The clustering model applied to the REF data Results of the clustering Conclusion 3 4 THE WAY FORWARD... 3 Figures Figure : Chronic disease rates by administrator (25 REF study)...2 Figure : Asthma rates by administrator (25 REF study)...22 Figure 3: Raw Asthma rates by cluster...29 Figure 4: Raw total CDL rates by cluster...29 Tables Table 9: Descriptive statistics for each cluster...28 REF shadow returns 2

4 Background Figure below is a graphic presentation of the number of chronic lives for each of the four participating administrators in the 25 REF Study 5. From the graph, it is clear that there are differences in the levels of chronic diseases between the four administrators. Figure : Chronic disease rates by administrator (25 REF study) Chronic Lives Rate per, Lives REF Study Rate per, Lives Under DH MS OMHC MHGra Total Given the differences in the number of chronic lives, it follows that there would be differences in the actual reported rates for the different CDL conditions. Figure shows Asthma rates by administrator. 5 Methodology for the Determination of the Risk Equalisation Fund Contribution Table [Base 25, Use ] RETAP Recommendations Report No. 9, available at: ecommendation_.pdf REF shadow returns 2

5 Figure : Asthma rates by administrator (25 REF study) Asthma 6. Rate per, Lives REF Study Under Rate per, Lives Expected 22 DH MS OMHC MHGr MHGr2 Within each administrator, there is variance in respect of the level of REF risk factors between schemes and benefit options. The previous REF report indicated that the industry average is an inappropriate benchmark for the analysis of REF returns. (See section 2..3, page 2, in the main report). RETAP agreed to this approach and supported the CMS to research alternative methods on clustering to group similar benefit options together and calculate a rate table for each cluster consisting of a group of similar benefit options. A scheme specific rate table could then be based on the respective cluster rate tables. 2 Introduction to cluster analysis Cluster analysis techniques encompass a number of different algorithms and methods for grouping similar objects in categories. A general question facing researchers in many areas of inquiry is how to organize data into meaningful structures, that is, to develop taxonomies. Cluster analysis is an exploratory data analysis tool, which aims at sorting objects into groups in such a manner that the degree of association between two objects is maximal if they belong to the same group and minimal otherwise. With clustering, there is no dependant variable, unlike regression analysis or logistic regression. Clustering, also known as unsupervised classification is classification with an unknown target. That is, the class of each case is unknown. Furthermore, the total number of classes is unknown. The aim is to segment the cases into disjoint classes that are homogeneous with respect to the inputs. REF shadow returns 22

6 2. Clustering methods There are literally hundreds of different algorithms that could be used to form the clusters. The choice of the algorithm and the choice of input variables can lead to different cluster solutions. The most common subset of algorithms is Hierarchical Cluster Analysis and Non-hierarchical (K-Means Cluster Analysis). This section introduces the basic concepts. 2.. Hierarchical Cluster Analysis The hierarchical procedure either starts with one cluster containing all individuals from which smaller clusters are formed by division, or starts with all individuals in separate clusters, which are then united step by step K-Means Cluster Analysis The K-Means algorithm is one of the most commonly used clustering algorithms. The K in its name refers to the fact that the algorithm looks for a fixed number of clusters, which are defined in terms of proximity of data points to each other. Each observation is assigned to the nearest seed (by Euclidean distance) to form a cluster. The seeds are then replaced by the means of the temporary clusters. The process continues until there are no further changes in the cluster means. The adjoining figure explains this for two variables, x and x 2. The steps are: Step : Select k observations randomly (seeds). Step 2: Assign all the remaining observations to the closest seed to form the original clusters. Step 3: Calculate the centroids of the clusters. The centroids become the seeds for the next iteration of the algorithm. Step 4: Re-assign observations to the closest seed (iterative process). REF shadow returns 23

7 The process of assigning points to a cluster and then recalculating centroids continues until the cluster boundaries stop changing. In practice, the K-Means algorithm usually finds a set of stable clusters after a few dozen iterations. 2.2 Similarity of - and distance between clusters Clustering methods depend on a measure of distance or similarity between points. Distance measures are preferred for quantitative data, and similarity measures for qualitative data. Different distance metrics used in clustering can give different types of clusters. The most widely used metric is Euclidean distance (L norm). The Euclidean distance between two points is the length of the straight line that joins them. Clusters formed using Euclidean distance tends to be spherical in nature. One problem with Euclidean distances is that they are greatly influenced by variables that have the largest values (outliers). One way around this problem is to standardise the variables. Euclidean Distance ( U2, V2 ) The statistical software packages frequently use the following three methods for calculating cluster distance: ( U, V ) L = ( U U ) + ( V V ) Average The distance between two clusters is the average distance between pairs of observations, one in each cluster Centroid The distance between two clusters is the Euclidean distance between their centroids or means Ward Cluster membership is assessed by calculating the total sum of squared deviations from the mean of a cluster. REF shadow returns 24

8 2.3 Evaluation criteria applied in evaluating clustering models There is no perfect way to determine the number of clusters, however there is a number of statistics than can be analysed to help with the process. These are the Pseudo-F statistic, the Cubic Clustering Criteria (CCC) and the Approximate Overall R-Squared Pseudo-F statistic This measure is similar to R 2, adjusted for the number of clusters. The larger the number of clusters the smaller the Pseudo-F value Cubic Clustering criterion A maximum CCC-value of larger than 2 indicates a meaningful cluster analysis. Values of CCC between and 2 should be interpreted carefully Approximate Overall R 2 Measure the between group deviance (variation) versus the overall deviance. A value close to is indicative of a good cluster structure. Increasing the number of clusters will increase the value of R 2. The three can be used together to identify the number of clusters. The trend to look for is where the Pseudo-F statistic and CCC peak and where the R-Squared tapers off. The objective of clustering is to minimize the variation within a cluster and to maximize the distance between clusters. 3 Application of clustering techniques to benefit options 3. The data In order to perform a clustering, reliable and appropriate data is required. The most suitable data is the 25 REF Study 6 dataset and the statutory returns data. The 6 Methodology for the Determination of the Risk Equalisation Fund Contribution Table [Base 25, Use ] RETAP Recommendations Report No. 9, available at: REF shadow returns 25

9 25 REF Study data includes prevalence and count data for 49 benefit options for 4,2 million lives. Only 2 of the 49 options were active in. If the 25 REF Study was used to do the clustering, more than 2 options will not be classified. The statutory returns data of were therefore used to do the clustering. The outcome (clustering of each option) is then linked back to the 25 REF Study data to calculate the actual rate tables for each cluster. The REF unit and RETAP considered the following variables in the clustering of the benefit options: Average age of the beneficiaries Proportion of beneficiaries above 65 years (Assumed pensioner ratio) Dependant ratio (Beneficiaries/Members) Proportion female lives Gross contribution per beneficiary per month Risk contribution per beneficiary per month Gross claims cost per beneficiary per month Risk claim premium ratio Risk claims cost per beneficiary (Jan-Sep ) Proportion of beneficiaries per age band: Under, to 4, 5 to 9 and subsequent five year bands up to 85, then 85 Plus Open: Open or restricted scheme Benefit design classification (used as dummy variables; / variables): High, Medium, Low, Hybrid, Capitation, Mixed 3.2 Clustering variables The input variables were discussed at the REF steering committee at RETAP technical committee meetings. Suggestions were made that CMS must include income, ethnicity distribution, region, type of managed care contract, etc., but unfortunately this data are not available at benefit option level. In the initial discussions it was concluded that the risk claims cost per beneficiary would be an important variable to describe the REF risk of a benefit option. The ecommendation_.pdf REF shadow returns 26

10 problem with the risk claims cost per beneficiary is that it is incomplete for and it is not cognisant to service provider arrangements (capitation, private versus provincial hospitals, etc.) or other measures to effectively of manage the risk of a benefit option or scheme. It is advised that all rand amount variables and ratio amount variables be used with caution in any risk classification model. 3.3 The clustering model applied to the REF data In the evaluation of the Q and Q2 data, REF Analysts applied the K-Means clustering model based on only the risk claims cost pbpm for 26. For the evaluation of the full year s REF data submissions, the REF unit used the model as described below. The identified list of variables was used in several different combinations (after standardisation) in a number of different clustering models (Hierarchical, K-Means, and multi-stage models) by the CMS and by external consultants. Based on statistical performance, the best three cluster models were selected and the rates for each model were calculated and graphed. (All the selected models had three clusters.) The models were compared on the outcomes for each benefit option (high, medium, and low). The results were disappointing in terms of a consensus outcome. For 33 out of a possible 375 benefit options there was no consensus between the three models. The three models were then evaluated by looking at the shapes for each CDL. From this visual analysis, the best clustering model was then selected. Scheme specific feedback, benefit design and the previous clustering results (Quarters and 2 ) were also taken into consideration in the final classification of each benefit option (Consensus model). 3.4 Results of the clustering The results for the consensus model are summarised in Table 9 below. The table presents descriptive statistics for each cluster. The high-risk cluster includes only 5 benefit options with a total of beneficiaries. These are benefit options with old and sick beneficiaries with an average age of 55,6 years, and a risk claims cost of R pbpm for. The REF shadow returns 27

11 high-risk cluster is also dominated by options with proportionally more females than males and the options typically make are an operating loss. The low-risk cluster includes 45 benefit options with a total of beneficiaries. These benefit options generally have young and healthy beneficiaries with an average age of 29,3 years and a risk claims cost of R46.9. These options typically make an operating profit. The medium cluster has 42 benefit options with a total of beneficiaries. The medium cluster lies between the high and the low cluster. Table 9: Descriptive statistics for each cluster Cluster N Variable Mean Std Dev Minimum Maximum Sum High 5 Beneficiaries Low 45 Medium 42 Average Age 55,63,69 39,54 76,37 Pensioner ratio 43,22 26,58 9,5 9,5 Female ratio 58,42 5,79 5,64 7,72 Dependant ratio,74,28,27,3 Claim premium ratio 7,49 5,44 82,57 34,4 Risk claims cost 474,7 378,94 79, 238,9 pbpm () Risk claims cost 293,96 35,27 658, ,37 pbpm (26) Beneficiaries , 5 527, Average Age 29,28 2,97 23,32 34,24 Pensioner ratio 3,79 2,55,2 8,95 Female ratio 5,26 4,26 39,4 6,89 Dependant ratio,29,26,7,87 Claim premium ratio 8,78 5,6 52,49 24,37 Risk claims cost 46,9 58,6 74,57 787,43 pbpm () Risk claims cost 399,8 266,6 47,5 74,86 pbpm (26) Beneficiaries Average Age 36,94 4,67 28,9 46,34 Pensioner ratio 2,4 5,92 3,4 24,96 Female ratio 52,57 3,5 46,9 58,44 Dependant ratio,27,26,75 2,8 Claim premium ratio 97,,95 72,22 44,33 Risk claims cost pbpm () Risk claims cost pbpm (26) 8,82 94,83 276,72 28,25 72,62 27,4 226,2 94, REF shadow returns 28

12 Figure 2 and Figure 3 shows the Asthma and total CDL rates for each cluster respectively, including the average rates found in the 25 REF Study for comparison. Figure 2: Raw Asthma rates by cluster Figure 3: Raw total CDL rates by cluster From the graphs above and the descriptive statistics shown in Table 9 it is clear that there are differences between the theoretical clusters in terms of the beneficiary profiles and the rates. REF shadow returns 29

13 3.5 Conclusion The clustering results are disappointing insofar as clear-cut hard definitions for these different clusters were not found. This does however occur frequently in cluster analyses, since it is dependant on variable selection. A clustering approach is a supportive rather than a definitive technique in the grouping of benefit options. There is much overlapping between the clusters. (A certain benefit option was for example classified as a low risk option, but it is on the border between the low and medium clusters. If a different set of input variables is applied, it might be classified as a medium risk option.) Clustering is an unsupervised approach and does not necessarily yield a solution that is predictive of the most important impacting variables. There are differences between benefit options and schemes, but in order to categorise a scheme correctly it is important for CMS to classify benefit options correctly and to allow enough variation around the expected rates in the evaluation of the REF submissions. The clustering approach is an improvement on the one size fits all approach, but continuous work will be done to improve on this technique and to evaluate alternatives. 4 The way forward A RETAP technical committee agreed that the consensus model is the optimal model for the time being to evaluate the submissions, but that more research needs to be done on the topic of classification of benefit options. Several suggestions were made by RETAP and one of them was a Generalised Linear Model (GLM) approach to classify the benefit options. Discovery Health and Metropolitan Health Corporate offered their help in the identification of the best set of variables at benefit option level to describe the risk of a benefit option. The so-called best set of variables will be discussed at RETAP and the REF steering committee. If necessary, CMS may in future adjust their statutory return specification on data collection to gather the critical information that will enable CMS to classify a benefit option more accurately in future. REF shadow returns 3

14 Schemes are invited to give feedback to the CMS if they do not agree with the clustering of their benefit options. Note that the classification was done in relation to the risk of all the other benefit options in the industry. The classification results (high, medium, and low) per benefit option are published in the individual scheme-specific reports on the CMS website 7. 7 The CMS statutory returns portal is available at: Note that a username and password is required to access scheme-specific information REF shadow returns 3

15 Annexure C: Category definitions Table : Full description of Category definitions Category 8 Full description Group 3 L There are some concerns with the submission that needs to be addressed. The community rate may not be the correct values until all the concerns are addressed. Total CDL levels; or three of MAT, CMY, HYP, IHD, HIV, CC SD; are on average 2 to 3 standard deviations below the scheme-specific expected rate, or; The CDL levels are even lower than above, but the office has collateral evidence that substantiates these low levels as a true reflection of the scheme's risk profile. 3 There are some concerns with the submission that needs to be addressed. The community rate may not be the correct values until all the concerns are addressed. 3 H There are some concerns with the submission that needs to be addressed. The community rate may not be the correct values until all the concerns are addressed. Total CDL levels; or three of MAT, CMY, HYP, IHD, HIV, CC SD; are on average 2 to 3 standard deviations above the scheme-specific expected rate, or; The CDL levels are even higher than above, but the office has collateral evidence that substantiates these low levels as a true reflection of the scheme's risk profile. 4 Substantially more REF beneficiaries than SR. 5 No REF data, or many beneficiaries missing. Fair data Poor data 6 LOW Total CDL / 3 LOW of MAT, CMY, HYP, IHD, HIV, CC2. No collateral evidence & >3 SD. 7 HIGH Total CDL / 3 HIGH of MAT, CMY, HYP, IHD, HIV, CC2. No collateral evidence & >3 SD. 8 Maternity data unlikely. 9 Combinations of the above, or other serious errors in submitted data, including but not limited to poor correlation between REF & SR data, unrealistic risk factor reporting that could not be classified in accordance with the other 8 categories, duplicate data submissions. CDL definitions applied poorly Poor data 8 Note that categories and 2, which were previously used to and identify good datasets with minor and no concerns respectively, has been discontinued. REF shadow returns 32

16 Annexure D: DIN Score methodology Contents BACKGROUND METHOD Data items used Degree of correlation between REF submissions and the statutory returns data in each age band REF risk factors Method applied to calculate scores Degree of correlation between REF submissions and the statutory returns data in each age band REF risk factor scores REF shadow returns 33

17 Background Medical schemes submit risk factor data in the form of REF grids with age, chronic diseases (CDLs), HIV, and maternity to the Council for Medical Schemes on a quarterly basis for the purpose of the REF shadow process. REF grids allow schemes to present their data in a consolidated format. For the accurate calculation of the scheme and industry community rates, it is important that medical schemes submit credible data. Despite the fact that a majority of scheme submit credible data, there are a number of schemes that continue to submit data with many errors. The office is continuously developing techniques for rapid evaluation of data submitted by medical schemes. One of the techniques is the application of Deviation from the Industry Norm (DIN) scores. The DIN scores use the principle of standard deviation to quantify the difference between the expected and submitted data. The DIN scores will allow analysts to flag schemes that require detailed investigation. DIN scores range from zero to ten, with low scores (<3) reflecting data with minor problems and ten () an indication of very bad data or no data submitted at all. The risk factor data that are reported at rates that are significantly lower or higher than expected will attract high DIN scores, indicating suspicious data. Deviation from the Industry Norm score is calculated for each scheme as an aid to estimate the quality of data submitted. The scores are based on the scheme s CDL data deviations from the scheme specific expected CDL profile for that scheme and the statutory returns data for the same period. The expected count of chronic diseases, maternity, and HIV cases per thousand beneficiaries is published with the REF Contribution Table on the CMS website. These rates are specific for high, medium, or low risk benefit options (See Annexure B, page 2, for details on the clustering of benefit options). These rate tables are used as a benchmark against which DIN scores for the REF risk factors are calculated. The number of beneficiaries in the REF grids is compared to the statutory returns data to measure, by age band, the accuracy of the submitted data. REF shadow returns 34

18 2 Method 2. Data items used DIN scores are calculated for each of the following data items: 2.. Degree of correlation between REF submissions and the statutory returns data in each band (a) (b) (c) Correlation between statutory returns data and the REF submissions by age band and total beneficiaries Correlation between statutory returns data and the REF submissions in the under year age category. Correlation between statutory returns data and the REF submissions in the age 85 plus age category 2..2 REF risk factors (a) CDL (25, MAT and HIV) in line with known prevalence and age-profile information (b) Aggregate REF risk factors score 2.2 Method applied to calculate scores 2.2. Degree of correlation between REF submissions and the statutory returns data in each age band Total data submitted score Calculate percentage deviation of the proportion of each age band to total beneficiaries between REF and SR returns o Abs( (Age(REF Agei)/ Total (REF beneficiaries) )/(Age (SR Agei) / Total( SR beneficiaries))* ) Calculate the cube root of the sum of percentage deviations for all age bands...(a) Calculate the cube root of the percentage deviation between the total beneficiaries of REF and SR returns o (Abs(Total (REF beneficiaries) /Total (SR beneficiaries) )* )) /3...(B) Data Submitted Score = (A + B)/2 REF shadow returns 35

19 Under age band score Calculate percentage deviation of the proportion of under s to total beneficiaries between REF and SR returns o Abs((Age (REF Under ) / Total (REF beneficiaries) )/(Age (SR Under ) / Total (SR beneficiaries))* ) The under score is calculated by finding the cube root of the percent deviation of the proportion of under s to total beneficiaries between REF and SR returns (maximum score = ) plus age band score Calculate percentage deviation of the proportion of 85 plus to total beneficiaries between REF and SR returns o Abs((Age (REF 85 plus) / Total (REF beneficiaries) ) / (Age (SR 85 plus) / Total (SR beneficiaries))* ) The 85 plus score is calculated by finding the cube root of the percent deviation of the proportion of under s to total beneficiaries between REF and SR returns (maximum score = ) REF risk factor scores CDL conditions score Required fields/variables: Expected; Actual for each age band Use published expected count rate tables to calculate scheme specific expected count Calculate the standard deviation (SD) of the expected proportion of CDL events (e.g. ADS) per age band: o SD = (square root((p*(-p*))/n) where p*=(x+2)/(n+4) and x = Expected CDL Age i ; n = Beneficiaries Age i Express the proportion SD as a numbers of lives with the CDL condition: o SD x Total Beneficiaries Express the difference (absolute difference) between the Actual and Expected number of CDL events in terms of the number of standard deviations per age band, and transform into bins as shown below: o ( Expected CDL Age i - Observed CDL Age i ) / (SD x Total Beneficiaries ) o Bins: : ; > 2: 2; >2 3: 3; >3 4: 4; >4 5: 5; >5 6: 6; >6 7: 7; >7 8: 8; >8 9: 9; >9: REF shadow returns 36

20 Multiply the binned values with the proportion of beneficiaries in each age band. The CDL DIN score is the sum these products in all the 9 age bands. o DIN Score (per CDL) = binned value age under x beneficiaries age under / total scheme beneficiaries binned value age 85 plus x beneficiaries age 85 plus / total scheme beneficiaries Aggregate REF risk factors DIN score The aggregate REF risk factor DIN score is calculated by computing the sum of the product for all REF risk factor (25 CDL, HIV, MAT) and the product of REF cost of the risk factor and proportion expected REF risk factor events: o Aggregate DIN Score = DIN CDL x REF Cost CDL x Total Expected DIN CDL / total scheme beneficiaries DIN CDL 25 x REF Cost CDL 25 x Total Expected DIN CDL 25 / total scheme beneficiaries + DIN MAT x REF Cost MAT x Total Expected DIN MAT / total scheme beneficiaries + DINHIV x REF Cost HIV x Total Expected DINHIV / total scheme beneficiaries REF shadow returns 37

21 Annexure E: REF submissions for, the categorisation thereof, and the potential financial impact of the REF Contents REF RETURNS SUBMITTED ASSESSMENT OF SUBMITTED DATA EVALUATION OF REF SUBMISSIONS BY ADMINISTRATOR THE POTENTIAL FINANCIAL IMPACT OF REF ON MEDICAL SCHEMES Introduction Analysis of the financial impact Sensitivity analysis 57 Figures Figure 4: Percentage of schemes with fair data...4 Figure 5: Percentage of schemes with fair data and poorly applied CDL definitions...42 Figure 6: Number of beneficiaries by payment band (December )...5 Figure 7: Number of beneficiaries by payment band (December, Alternative payment intervals)52 Figure 8: Scheme risk base on the Full table (December )...53 Figure 9: Industry community rate: With and without exclusions...57 Tables Table : REF and SR returns submitted for March Table 2: REF and SR returns submitted for June... 4 Table 3: REF and SR returns submitted for September... 4 Table 4: REF and SR returns submitted for December... 4 Table 5: Number of schemes by category and month Table 6: Scheme categories by administrator (March ) Table 7: Scheme categories by administrator (June ) Table 8: Scheme categories by administrator (September ) Table 9: Scheme categories by administrator (December ) Table 2: Risk rates by month Table 2: Frequency distribution of the number of schemes versus the scheme risk in intervals Table 22: Frequency distribution of the number of beneficiaries versus the scheme risk in intervals... 5 Table 23: Frequency distribution of the number of beneficiaries versus the scheme risk in intervals (Alternative intervals)... 5 Table 24 Frequency distribution of the number of schemes versus the scheme risk in intervals (Alternative payment intervals) Table 25: Scheme risk by category (December ) Table 26: Detailed list of scheme risks for December Table 27: Number of schemes excluded per month Table 28: Risk rates per month without Category 4, 5, 6, 7, 8, and 9 schemes REF shadow returns 38

22 REF returns submitted Table indicates that during March REF data for 2 schemes were included in the analysis, representing beneficiaries in the industry (99,35 percent of the total number of beneficiaries reported in the statutory returns). See Annexure C (page 32) for the definitions of the respective categories. Table : REF and SR returns submitted for March Category Number of Schemes Statutory Returns Beneficiaries in March Percentage REF grids of Total SR Submitted Beneficiaries REF Beneficiaries as % SR Beneficiaries 3 L 2 (%) , ,8 3 7 (53,33%) , ,75 3 H (9,7%) , , (,67%) 32 88, ,4 6 2 (,67%) , , (3,33%) , , (5,63%) , ,35 Total % ,35% The following schemes are exempted from PMBs and were not included in the analysis: Building and Construction Industry Medical Aid Fund Fishing Industry Medical Scheme (Fish-Med) Food Workers Medical Benefit Fund Golden Arrows Employees Medical Benefit Fund Gen-Health Medical Scheme is the only scheme that did not submit any REF data for and was classified as a Category 5 scheme (no data submitted, or there are many beneficiaries missing on the REF submission) for every month. Gen-Health Medical Scheme was included in the categorisation results, but not in any of the other analysis. Table 2 indicates that during June, REF data for 2 schemes were included in the analysis, representing beneficiaries in the industry (99,55 percent of the total number of beneficiaries reported in the statutory returns). REF shadow returns 39

23 Table 2: REF and SR returns submitted for June Decision Category Number of Schemes Statutory Returns Beneficiaries in June Percentage REF grids of Total SR Submitted Beneficiaries REF Beneficiaries as % SR Beneficiaries 3 L 2 (%) , , (5,67%) , ,85 3 H 3 (,83) , , 4 (,83%) 24 59, , (2,5%) 5 632, , (2,5%) , , (4,7%) , ,55 8 (,83%) , ,8 9 2 (6,67%) , ,9 Total % ,55 Table 3 indicates that during September REF data for 7 schemes were included in the analysis, representing beneficiaries in the industry (99,38 percent of the total number of beneficiaries reported in the statutory returns). Table 3: REF and SR returns submitted for September Decision Category Number of Schemes Statutory Returns Beneficiaries in September Percentage REF grids of Total SR Submitted Beneficiaries REF Beneficiaries as % SR Beneficiaries 3 L 9 (7,69%) , , (57,26%) , ,8 3 H 3 (,%) , , (,85%) , (,7%) , ,8 7 2 (,7%) , (9,66%) , ,44 Total % ,38 Table 4 below indicates that during December REF data for 7 schemes were included in the analysis, representing beneficiaries in the industry (99,9 percent of the total number of beneficiaries reported in the statutory returns). REF shadow returns 4

24 Table 4: REF and SR returns submitted for December Decision Category Number of Schemes Statutory Returns Beneficiaries in December Percentage REF grids of Total SR Submitted Beneficiaries REF Beneficiaries as % SR Beneficiaries 3 L 8 (6,84%) , , (59,83%) , ,77 3 H 5 (2,82%) , , (,7%) , , 6 (,85%) 6 69, , (,7%) , , (6,24%) , , Total % ,9 2 Assessment of submitted data During the year, there was a slight improvement in the number of schemes grouped as submitting fair data (Categories 3, 3 L, or 3 H). This trend is displayed in Figure 4. In January 76,67 percent of the schemes were assessed to have fair data. This percentage went up to almost 8 percent in December. The slight improvement may be the result of the feedback sessions that CMS had with some schemes. Figure 4: Percentage of schemes with fair data. Percentage Jan-7 Feb-7 Mar-7 Apr-7 May-7 Jun-7 Jul-7 Aug-7 Sep-7 Oct-7 Nov-7 Dec-7 Month In Figure 5 Categories 6 and 7 included; the respective percentages increased with approximately 2 to 3 percent per month above the results reported in Figure 4, except for April where the percentage went up with 7,5 percent. The percentage of fair data was generally lower for quarter 2 compared to the other quarters. REF shadow returns 4

25 Figure 5: Percentage of schemes with fair data and poorly applied CDL definitions Percentage Jan-7 Feb-7 Mar-7 Apr-7 May-7 Jun-7 Jul-7 Aug-7 Sep-7 Oct-7 Nov-7 Dec-7 Month REF shadow returns 42

26 Table 5: Number of schemes by category and month Period Category Frequency Row Pct Col Pct 3 L 3 3 H Total Jan-7 9,7 9,7 7 58,33 8,58 9,7 7,9 3 2,5, 2,67 8,33 4 3,33 9,52 9 5,83 8,9 2 Feb-7 9,7 9,7 Mar-7 2 Apr-7 9,7 9,7 May-7 2 Jun-7 2 Jul-7 9 7,63 7,5 Aug-7 9 7,63 7,5 Sep-7 9 7,69 7,5 Oct-7 8 6,84 6,67 Nov-7 8 6,84 6,67 Dec-7 8 6,84 6, ,33 8, ,33 8, ,5 7, ,7 7, ,67 7, ,47 8, ,47 8, ,26 8,2 7 59,83 8,58 7 6,68 8,7 7 59,83 8,58 9,7 7,9 9,7 7,9 3,83 8,5 2 7,84 3,83 8,5 3,2 8,5 3,2 8,5 3, 8,5 4,97 9,5 4,97 9,5 5 2,82 9,8,83 6,67,83 6,67 2,67 33,33,83 6,67,85 6,67 2,67 7,4 2,67 7,4 4 3,33 4,8 3 2,5, 3 2,5,,85 3,7,85 3,7,85 3,7 3 2,56, 2,7 7,4 2,7 7,4 2,67 8,33 2,67 8,33 4 3,33 6,67 3 2,5 2,5 3 2,5 2,5 2,69 8,33 2,69 8,33 2,7 8,33,85 4,7,85 4,7 4 3,33 9,52 4 3,33 9,52 5 4,7,9 5 4,7,9 5 4,7,9 3 2,54 7,4 2,69 4,76 2,7 4,76 3 2,56 7,4 3 2,56 7,4 2,7 4,76,83 9 5,83 8,9 9 5,83 8,9 9 5,83 8,9 8 5, 7,66 2 6,67 8,5 2 7,8 8, ,64 9, ,66 9,79 8 5,38 7,66 8 5,38 7,66 9 6,24 8,9 Total Between 8 (5.38%) and 23 (9,66%) schemes were classified as a Category 9 schemes. These submissions contain gross irregularities in more than one area (see Annexure C on page 32 for category definitions) Evaluation of REF submissions by administrator The number of schemes per administrator is counted for each category and the results are reported for the last month in each quarter in Table 6 to Table 9 on pages 44 to 47. REF shadow returns 43

27 Table 6: Scheme categories by administrator (March ) Administrator vs. Category Administrator Category Frequency Row Pct 3 L 3 3 H Total ALLCARE ADMINISTRATORS (PTY) LTD ,43 28,57 AMANZI HEALTH ADMINISTRATORS (PTY) LTD DISCOVERY HEALTH (PTY) LTD 9 9 ETERNITY PRIVATE HEALTH FUND ADMINISTRATORS (PTY) LTD FULL CIRCLE HEALTH (PTY) LTD 2 2 INGWE MED (PTY) LTD MEDSCHEME HOLDINGS (PTY) LTD , 5, METROPOLITAN HEALTH CORPORATE 3 6 (PTY) LTD 8,75 68,75 6,25 6,25 MPUMALANGA MANAGED HEALTH CARE (PTY) LTD MULTIMED HEALTHCARE ADMINISTRATORS (PTY) LTD OLD MUTUAL HEALTHCARE (PTY) LTD ,56 22,22,, PPS MEDICAL SCHEME ADMINISTRATORS (PTY) LTD PRIVATE HEALTH ADMINISTRATORS PROSPERITY HEALTH MANAGERS 2 (PTY) LTD 5 5 PROVIDENCE HEALTHCARE RISK 2 4 MANAGERS (PTY) LTD 25, 5 25, RESOLUTION ADMINISTRATORS (PTY) LTD ROWAN ANGEL (PTY) LTD SECHABA MEDICAL SOLUTIONS (PTY) 2 LTD 5 5 SEKMED MEDICAL AID ADMINISTRATORS (PTY) LTD SELF-ADMINISTERED ,25 3,25 8,75 6,25 37,5 SIGMA HEALTH FUND MANAGERS (PTY) LTD MOMENTUM MEDICAL SCHEME 8 2 ADMINISTRATORS (PTY) LTD 8 2 STATUS MEDICAL AID ADMINISTRATORS (PTY) LTD 5 37,5 2,5 THEBE YA BOPHELO HEALTHCARE ADMINISTRATORS (PTY) LTD V MEDICAL AID ADMINISTRATORS (PTY) LTD Total REF shadow returns 44

28 Table 7: Scheme categories by administrator (June ) Administrator vs. Category Administrator Category Frequency Row Pct 3 L 3 3 H Total ALLCARE ADMINISTRATORS (PTY) LTD 57,4 4,29 28,57 AMANZI HEALTH 2 ADMINISTRATORS (PTY) LTD 5 5 DISCOVERY HEALTH (PTY) LTD 9 8,82 9,9 9,9 ETERNITY PRIVATE HEALTH FUND ADMINISTRATORS (PTY) LTD FULL CIRCLE HEALTH (PTY) 2 2 LTD INGWE MED (PTY) LTD MEDSCHEME HOLDINGS (PTY) 7 2 LTD 5, 85, 5, 5, METROPOLITAN HEALTH 3 6 CORPORATE (PTY) LTD 8,75 68,75 6,25 6,25 MPUMALANGA MANAGED HEALTH CARE (PTY) LTD MULTIMED HEALTHCARE ADMINISTRATORS (PTY) LTD OLD MUTUAL HEALTHCARE (PTY) LTD 44,44 33,33,, PPS MEDICAL SCHEME ADMINISTRATORS (PTY) LTD PRIVATE HEALTH ADMINISTRATORS PROSPERITY HEALTH 2 MANAGERS (PTY) LTD 5 5 PROVIDENCE HEALTHCARE RISK MANAGERS (PTY) LTD ROWAN ANGEL (PTY) LTD RESOLUTION ADMINISTRATORS (PTY) LTD SECHABA MEDICAL 2 SOLUTIONS (PTY) LTD 5 5 SELF-ADMINISTERED ,67 26,67 2 6,67 3,33 26,67 SIGMA HEALTH FUND MANAGERS (PTY) LTD MOMENTUM MEDICAL SCHEME ADMINISTRATORS (PTY) LTD STATUS MEDICAL AID ADMINISTRATORS (PTY) LTD THEBE YA BOPHELO HEALTHCARE ADMINISTRATORS (PTY) LTD 5 25, 2,5 2,5 V MEDICAL AID ADMINISTRATORS (PTY) LTD Total REF shadow returns 45

29 Table 8: Scheme categories by administrator (September ) Administrator vs. Category Administrator Category Frequency Row Pct 3 L 3 3 H Total ALLCARE ADMINISTRATORS (PTY) LTD ,4 42,86 AMANZI HEALTH ADMINISTRATORS (PTY) LTD DISCOVERY HEALTH (PTY) LTD 9 9 ETERNITY PRIVATE HEALTH FUND ADMINISTRATORS (PTY) LTD FULL CIRCLE HEALTH (PTY) LTD 2 2 INGWE MED (PTY) LTD MEDSCHEME HOLDINGS (PTY) LTD ,95,53,53 METROPOLITAN HEALTH CORPORATE 3 6 (PTY) LTD 8,75 68,75 6,25 6,25 MPUMALANGA MANAGED HEALTH CARE (PTY) LTD MULTIMED HEALTHCARE ADMINISTRATORS (PTY) LTD OLD MUTUAL HEALTHCARE (PTY) LTD ,5 2,5 PPS MEDICAL SCHEME ADMINISTRATORS (PTY) LTD PRIVATE HEALTH ADMINISTRATORS PROSPERITY HEALTH MANAGERS (PTY) 2 LTD 5 5 PROVIDENCE HEALTHCARE RISK MANAGERS (PTY) LTD ROWAN ANGEL (PTY) LTD RESOLUTION ADMINISTRATORS (PTY) LTD SECHABA MEDICAL SOLUTIONS (PTY) LTD SELF-ADMINISTERED , ,67 6,67 4 SIGMA HEALTH FUND MANAGERS (PTY) LTD MOMENTUM MEDICAL SCHEME 8 2 ADMINISTRATORS (PTY) LTD 8 2 STATUS MEDICAL AID ADMINISTRATORS (PTY) LTD 37,5 5 2,5 THEBE YA BOPHELO HEALTHCARE ADMINISTRATORS (PTY) LTD V MEDICAL AID ADMINISTRATORS (PTY) LTD Total REF shadow returns 46

30 Table 9: Scheme categories by administrator (December ) Administrator vs. Category Administrator Category Frequency Row Pct 3 L 3 3 H Total ALLCARE ADMINISTRATORS (PTY) LTD ,4 4,29 28,57 DISCOVERY HEALTH (PTY) LTD ETERNITY PRIVATE HEALTH FUND ADMINISTRATORS (PTY) LTD FULL CIRCLE HEALTH (PTY) LTD HDS Medical (Pty) Ltd HWH INTEGRATED RISK MANAGEMENT D/O TRIANGULAR HEALTH (PTY) LTD INGWE MED (PTY) LTD MEDSCHEME HOLDINGS (PTY) LTD ,2,53 5,26 METROPOLITAN HEALTH CORPORATE (PTY) LTD 2,5 75, 6,25 6,25 MOMENTUM MEDICAL SCHEME 7 3 ADMINISTRATORS (PTY) LTD 7 3 MULTIMED HEALTHCARE ADMINISTRATORS (PTY) LTD OLD MUTUAL HEALTHCARE (PTY) LTD , 25, PPS MEDICAL SCHEME ADMINISTRATORS (PTY) LTD PRIVATE HEALTH ADMINISTRATORS PROSPERITY HEALTH MANAGERS (PTY) 2 LTD 5 5 PROVIDENCE HEALTHCARE RISK MANAGERS (PTY) LTD ROWAN ANGEL (PTY) LTD RESOLUTION ADMINISTRATORS (PTY) LTD SECHABA MEDICAL SOLUTIONS (PTY) LTD SELF-ADMINISTERED ,25 37,5 8,75 6,25 6,25 25, SIGMA HEALTH FUND MANAGERS (PTY) LTD STATUS MEDICAL AID ADMINISTRATORS (PTY) LTD 37,5 5 2,5 THEBE YA BOPHELO HEALTHCARE ADMINISTRATORS (PTY) LTD V MEDICAL AID ADMINISTRATORS (PTY) LTD Total REF shadow returns 47

31 4 The potential financial impact of REF on medical schemes 4. Introduction The scheme s risk (industry community rate scheme community rate) was calculated individually for each scheme for March, June, September, and December based on the Full contribution table. Initially all the schemes were included in the calculation of the industry community rate and then the Category 4, 5, 6, 7, 8 and 9 schemes were excluded in the analysis. 4.2 Analysis of the financial impact One hundred and nineteen schemes were included in the analysis for March and June, while one hundred and sixteen were included for September and December 9. Basic statistics are shown for each of the four months in Table 2. Contrary to the deviation in 26, the scheme risk was stable from quarter to quarter during. For March, the scheme risk varies from R765,27 to R8,43. This means that if these datasets are a true reflection of the respective schemes risk, the highest risk scheme may receive R765,27 per beneficiary from REF and the lowest risk scheme may pay R8,43 per beneficiary to REF. For December, the scheme risk ranges from R8,82 to R99,43. Table 2: Risk rates by month Statistic Full Contribution Table (Amount in rand) March June September December Industry community rate 258,74 259,25 257,99 26,36 Minimum risk rate -765,27-775,43-797,73-8,82 Maximum risk rate 8,43 3,25,66 99,43 Standard deviation 4, 5,95 5, 5,78 Fifty schemes (43,%) were net contributors in December, but these fifty schemes presents (72,32%) beneficiaries. 9 Note that Gen-health medical scheme, although included in the categorisation, is excluded form all community rate analysis because they have not submitted any REF returns. REF shadow returns 48

32 Table 2: Frequency distribution of the number of schemes versus the scheme risk in intervals Scheme risk March June September December Schemes % Schemes % Schemes % Schemes % Pay: R to 9 5,97 8 5,3 8 5,5 2 7,2 R25, pbpm 2 4 Pay: R25, to R5 pbpm 9,24 5 2,6 6 3,7 9 3,2 Pay: R5, to R75, pbpm 2,8 3,92 9 7,76 2,3 4 Pay: R75, to R pbpm 9 7,56 4 3,36 6 5,7 5 4,3 Pay: R,,84 2,68, to R25, pbpm Pay: R25, to R5 pbpm Pay: More than R5 pbpm Sub-total 52 43, ,7 5 43, 5 43, Receive: R, 8 5,3 6 3,45 9 6,3 4 2, to R25, 8 7 pbpm Receive: R25, to R5 pbpm Receive: R5, to R75, pbpm Receive: R75, to R pbpm Receive: R, to R25, pbpm Receive: R25, to R5 pbpm Receive: More than R5 pbpm 7 4,29 4,76 4 2, 7 6 3, ,88 8 6,72 8 6,9 4 2, 7 7 5,88 9,24 8 6,9 3 2,59 5 4,2 5 4,2 7 6,3 8 6,9 6 5,4 5 4,2 2,72 3 2,59 7 5,88 8 6,72 8 6,9 8 6,9 Total REF shadow returns 49

33 Table 22: Frequency distribution of the number of beneficiaries versus the scheme risk in intervals Scheme risk March June September December Beneficia % Beneficia % Beneficia % Benefic % ries ries ries iaries Pay: R to R25, pbpm , , , ,86 Pay: R25, to , , , ,88 R5 pbpm Pay: R5, to , , , ,23 R75, pbpm 839 Pay: R75, to , , , ,35 R pbpm Pay: R, to 7 799, , ,3 - - R25, pbpm Pay: R25, to R5 pbpm Pay: More than R5 pbpm Sub-total net , , , ,32 payers 889 Receive: R, to , , , ,5 R25, pbpm Receive: R25, to , , , ,95 R5 pbpm Receive: R5, to , , , ,4 R75, pbpm Receive: R75, to , , , , R pbpm Receive: R, to 3 952, , , 3 388,4 R25, pbpm Receive: R25, to , , , ,88 R5 pbpm Receive: More than R5 pbpm 398 5, , , ,8 Subtotal net 2 42 recipients , , , ,68 Total The financial impact by payment band on the beneficiaries is illustrated in Figure 6 (page 5) for December. Three hundred fifty four thousand and four hundred and ninety three (4,8%) beneficiaries may receive R5 or more from REF and 99 8 (,35%) may pay in between R75, and R pbpm. (Theoretically, more than 7% beneficiaries may be net payers into REF.) REF shadow returns 5

34 Figure 6: Number of beneficiaries by payment band (December ) Full table Number of beneficiaries 3,5 3, 2,5 2,,5, 5 Pay: More than R5 PBPM Pay: R25, to R5 PBPM Pay: R, to R25, PBPM Pay: R75, to R PBPM Pay: R5, to R75, PBPM Pay: R25, to R5 PBPM Pay: R to R25, PBPM Receive: R, to R25, PBPM Receive: R25, to R5 PBPM Receive: R5, to R75, PBPM Receive: R75, to R PBPM Receive: R, to R25, PBPM Receive: R25, to R5 PBPM Receive: More than R5 PBPM The payments are grouped differently for December in Table 23, Table 24 and Figure 7 below. If we assume that payments less than R5 are not significant, then we could conclude that REF will have no or little effect on approximately 27 percent of the beneficiaries in the industry. These 27 percent beneficiaries are in 5 different schemes. Only 45,3 percent will then be net payers compared to the 72,32 percent shown in Table 22. Table 23: Frequency distribution of the number of beneficiaries versus the scheme risk in intervals (Alternative intervals) Scheme risk (December ) Number of beneficiaries Percent (%) Cumulative number of beneficiaries Cumulative percent (%) Pay more than R75 pbpm 99 8, ,35 Pay between R4 and R75 pbpm , ,8 Pay between R5 and R4 pbpm , ,3 Paying or receiving less than R5 pbpm , ,32 Receive between R5 and R4 pbpm , ,7 Receive between R4 and R75 pbpm , ,9 Receive more than R75 pbpm , REF shadow returns 5

35 Figure 7: Number of beneficiaries by payment band (December, Alternative payment intervals) Full table Number of beneficiaries 2,5 2,,5, Pay more than R75, PBPM Pay between R4 and R75, PBPM Pay between R5 and R4 PBPM Paying or Receiving less than R5 PBPM Receive between R5 and R4 PBPM Receive between R4 and R75, PBPM Receive more than R75 PBPM Table 24 Frequency distribution of the number of schemes versus the scheme risk in intervals (Alternative payment intervals) Scheme risk (December Number of schemes Percent (%) Cumulative number of schemes Cumulative percent (%) Pay more than R75 pbpm 5 4,3 5 4,3 Pay between R4 and R75 pbpm 5 2,93 2 7,24 Pay between R5 and R4 pbpm 25 2, ,79 Paying or receiving less than R5 pbpm 5 4,3 5 43, Receive between R5 and R4 pbpm 25 2, ,66 Receive between R4 and R75 pbpm 9 6, ,3 Receive more than R75 pbpm 22 8,97 6 Figure 8 below illustrates the variation in the scheme risk based on the full contribution table for December 2. Based on the submitted data there is one scheme that will receive R8.82 per beneficiary for December. This is a small scheme with less than 5 beneficiaries and REF analysts classified it as Category 3. The maximum net payer for December (R99.43 per beneficiary) is a scheme with between and 3 beneficiaries and the scheme was classified as a Category 3 scheme by REF analysts. 2 CMS Approved contribution tables for Approved%2REF%2Contribution%2tables%2for%2.xls REF shadow returns 52