For various ranges of $a$ and $b$ the integrand looks like a bell curve with a steeper slope around small values of $u$. The integrand rises as $u$ approaches $\tfrac{1}{4}$ and reaches very small values around $u=\tfrac{1}{2}$.
On approximating $\left(\dfrac{u}{1-u}\right)^{2b-1}$ by its expansion around the point $u=\tfrac{1}{4}$, a rough approximation to the maximum, I get, to the first order in $u-\tfrac{1}{4}$,
$$\left(\dfrac{u}{1-u}\right)^{2b-1}\approx
3/(3^b)^2 + ((32 b - 16) (u - 1/4))/(3^b)^2.$$ When $b=\tfrac{1}{2}$, this term reduces to unity, and the integral reduces to the form given by gpmath.
Inserting this expression into the integrand and using Wolfram Alpha produces the result
$$\int \exp\left\{(-a/(2 (-1 + 2 u))\right\} \left\{3/(3^b)^2 + ((32 b - 16) (u - 1/4))/(3^b)^2\right\}\,{\rm d}u = (\tfrac{1}{4} 9^{-b} (2 (2 u - 1) e^{a/(2 - 4 u)} (a (2 - 4 b) + 8 (2 b - 1) u + 3) - a (a (4 b - 2) - 8 b + 1) \text{Ei}(a/(2 - 4 u))) + \text{constant}.$$
Differencing this relation for $u=\tfrac{1}{2}$ and $u=0$ would give a rough approximation to the integral. More terms in the Taylor expansion about the point $u=\tfrac{1}{4}$ could be taken, but the algebra increases exponentially, although I tried terms up to $\left(u-\tfrac{1}{4}\right)^3$ with no problems.