The effects of the Norwegian Coordination Reform on the use of rehabilitation services: panel data analyses of service use, 2010 to 2013

[1] provides an English summary of the reform.

As of May 2014, the municipalities employed 5.9 % of GPs in the Medical Associations database. Thus, most GPs are contractors.

These are the preliminaries of a possible counterfactual argument, stating that an expansion of municipal services would have occurred had MCF been more comprehensive (which we, of course, do not observe). Moreover, the low cost-sharing rate (20 % of specialist service costs paid by municipalities) is usually motivated by it being an offset to the budgetary risks (e.g. in the face of demographic change) of the smaller municipalities (compared with the larger hospital catchment areas). Nevertheless, the argument is not very plausible. Even with a higher cost-sharing rate, the consequences, apart from heavy strains on local authorities’ revenues, could very well be a heavier (short-term) reliance on (readily available) specialist capabilities and a weaker reliance on local capabilities (i.e. costs of a “sunken” character going into “shaky” investments in “longer-term” local services). In other words, the argument may simply be that municipalities in Norway are too small for incentive schemes such as the MCF to work. We return to this point in the discussion.

The DRGs are DRG-462A: Complex rehabilitation, DRG-462B: Ordinary rehabilitation, DRG-462C: Other rehabilitation, DRG-462O: Unspecified rehabilitation, DRG-932O: Policlinic rehabilitation, and DRG-998O: Group-based patient recovery. The choice of DRGs is based on a presentation given by Vidar Halsteinli for the Central Norway Regional Health Authority 29.09.2010.

Nominal DRG prices (for the average DRG) in the 2010–2013 period are NOK 35,964 (2010), NOK 36,968 (2011), NOK 38,209 (2012), and NOK 39,447 (2013) (Norwegian Directory of Health).

A standard measure for “municipal rehabilitation service” production is man-hours worked by physio- and ergotherapists and psychiatric nurses as this “is assumed to be closely related to rehabilitation”. In other words, the measure is coarser than analogues at the specialist and private level (where diagnoses, tasks, and procedures are better defined and priced). Moreover, records are not complete for a large proportion of the municipalities prior to 2013. The measure that we utilize therefore relies on the two former occupational groups (for which records are the most complete).

This constitutes 96 % of the 428 Norwegian municipalities. Missing values are most often associated with incomplete municipal reporting to KOSTRA.

Even though these results are weighted by municipal population; a point to which we return below.

Norwegian municipalities vary in population size from just above 200 to well over 600,000 (Oslo), and the median and mean population sizes stand at around 4500 and 12,000 respectively (2012).

These are the exact predictions for relative changes, i.e. exp(α)-1. For example, for the estimate of the α₂₀₁₂ effect this is exp(-0.15)-1 = 0.14. We report such exact effects in the text.

For example, for specialist services the estimate of α ₂₀₁₂ -α ₂₀₁₁  = -0.15-(-0.03) = -0.12 is significant at the p = 0.00 level, as are the analoguos differences for the other service types.

Although this pattern disappears with other specifictions (model variants (B) and (C); see below).

First stage results for two-stage least square estimations are presented in Additional file 2: Table S2.

For example, average (log) specialist service levels change by α ₂₀₁₂ -α ₂₀₁₁ , estimated at -0.11-(-0.02) = -0.09 from 2011 to 2012 (z = 2.30, p = 0.022). In the case of municipal rehabilitation services there are no signficant differences between years whatsoever.

In analyses in Table 3 we also cluster standard errors at the municipality level.

Although not very substantial, 2012–2013 changes, estimated by jointly testing α₀ + α ₂₀₁₃ , are significant at the 5 % level for specialist and private services in model variant (D).

Implemented by adding interaction terms between the ΔNeeds and ΔYear variables to equation (2). Results for interaction terms are shown in Additional file 3: Table S3.

For other service types (private and municipal), needs effects are alternately positive or negative, and never significant.

Conveniently, the additive α estimates may be interpreted as effects when the intreacted variable, the needs variables, are zero. In other words, α estimates are Year effects in the absence of needs changes.